Systems and methods for analyzing blood glucose and changing human behavior

This application claims the priority benefit of U.S. Provisional Patent Application Ser. No. 63/670,585, filed Jul. 12, 2024, which is herein incorporated by reference in its entirety.

All publications and patent applications mentioned in this specification are herein incorporated by reference in their entirety, as if each individual publication or patent application was specifically and individually indicated to be incorporated by reference in its entirety.

This disclosure relates generally to the field of general wellness, and more specifically to the field of diabetes management and related human behavior. Described herein are systems and methods that utilize large language models and/or generative artificial intelligence to change human behavior associated with managing health and diabetes.

The prevalence of diabetes, including type 2 diabetes (T2D), type 1 diabetes (TID), and prediabetes will increase to more than 54.9 million Americans by 2030 and is projected to cross half a billion people worldwide by 2040. Before developing T2D, most people have prediabetes where blood sugar values are higher than a predefined normal, but not high enough yet for a diabetes diagnosis. While the prevalence of such conditions is predicted to increase substantially in the near future, proper blood glucose management can effectively reverse this condition, thus improving the health outcomes of these individuals. Several aspects of the etiology of T2D suggest that both high-risk and whole-population approaches are helpful to make a major impact on trends in the diabetes epidemic.

Self-monitoring of blood glucose (SMBG) is a helpful disease management component for improving patient/device user self-management and guiding clinicians in adjusting medications. SMBG data is used to improve glycemic control and identify hypoglycemic events, and also persistent hyperglycemia. The clinical utility of SMBG is dependent upon patients understanding the proper technique, being adherent to testing, and correctly interpreting the results. In addition, clinical utility comes from patients sharing their testing results with providers, and providers acting on that data to make treatment decisions.

In some aspects, the techniques described herein relate to a computer-implemented method for managing behavior in a subject, the method including: monitoring a plurality of behavioral events, each event being correlated to a timestamp within a time period; identifying a pattern in the plurality of behavioral events; generating, based on the identified pattern, personalized content for the subject; generating one or more nudges for the subject based on the personalized content, wherein the one or more nudges include prompts or suggestions aimed at encouraging adherence to executing suggested behavior changes provided in the personalized content; and causing presentation of the personalized content and the one or more nudges to the subject.

In some aspects, the techniques described herein relate to a computer-implemented method, wherein: at least one of the plurality of behavioral events includes a last consumed meal; the personalized content includes feedback on the last consumed meal.

In some aspects, the techniques described herein relate to a computer-implemented method, wherein the feedback includes suggested timing for a next meal and educational content, the educational content including a meal plan for the subject that minimizes glucose spikes over a second time period.

In some aspects, the techniques described herein relate to a computer-implemented method, wherein the personalized content includes one or more of: dietary recommendations, exercise suggestions, and lifestyle modifications.

In some aspects, the techniques described herein relate to a computer-implemented method, wherein the method further includes: monitoring a behavior or input responsive to the presentation of the one or more nudges; iteratively repeating the monitoring of behavior exhibited by the subject; in response to detecting additional behavioral events, identifying an additional pattern in the additional behavioral events; updating the personalized content according to the additional behavioral events based on the identified additional pattern; generating an updated nudge corresponding to the updated personalized content; and causing presentation of the updated nudge and the updated personalized content to the subject.

In some aspects, the techniques described herein relate to a computer-implemented method, wherein the monitoring of the plurality of behavioral events is performed at least in part by an artificial intelligence computer agent configured to use data generated by the monitoring and a plurality of template guides to generate the one or more nudges.

In some aspects, the techniques described herein relate to a computer-implemented method, wherein the method further includes: adapting the personalized content based on contextual information associated with the subject, wherein the contextual information includes at least one of: a current location, a time of day, one or more upcoming events, or one or more recent activities.

In some aspects, the techniques described herein relate to a computer-implemented method, wherein the method further includes: determining, based on the plurality of behavioral events and the monitoring, a nudge mode from a plurality of nudge modes, a historical adherence to previous nudges or recommendations generated for the subject; and generating the one or more nudges according to the determined nudge mode.

In some aspects, the techniques described herein relate to a computer-implemented method, wherein the plurality of nudge modes includes a light mode, a regular mode, and a heavy mode.

In some aspects, the techniques described herein relate to a computer-implemented method, wherein the light mode is activated during at least one of: holidays, travel, postpartum periods, short-term illnesses, or user-designated cheat days.

In some aspects, the techniques described herein relate to a computer-implemented method, wherein the heavy mode is activated during at least one of: periods following discontinuation of one or more medications or when the subject sets aggressive health goals.

In some aspects, the techniques described herein relate to a computer-implemented method, wherein: monitoring the plurality of behavioral events includes monitoring or receiving blood glucose data for the subject; identifying the pattern in the plurality of behavioral events includes identifying a glucose time in range for the subject; and selecting, based on the glucose time in range, one or more times in which to present the one or more nudges to the subject.

In some aspects, the techniques described herein relate to a computer-implemented method, wherein the personalized content includes a grocery shopping list curated according to the identified pattern, and the method further includes automatically placing a plurality of items on the grocery shopping list in an online shopping cart or automatically placing an order of the plurality of items on the grocery shopping list.

In some aspects, the techniques described herein relate to a computer-implemented method, wherein the personalized content includes a plurality of restaurant menu suggestions selected according to the identified pattern, and food data corresponding to each of the plurality of restaurant menu suggestions, the food data indicating one or more of: calorie content, sugar content, carbohydrate content, and at least one reason for selecting the respective restaurant menu suggestions.

In some aspects, the techniques described herein relate to a computer-implemented method, wherein the computer-implemented method is performed by an artificial intelligence health coach configured to share data with a buddy computing device according to permissions associated with the subject, the shared data including one or more of: a blood glucose estimation, identified behavioral patterns, a status update corresponding to the subject, and a summary corresponding to the subject.

In some aspects, the operations described herein are performed by an artificial intelligence computer agent configured to use data generated by monitoring the behavior of the subject and a plurality of template guides to generate the one or more nudges to coach the subject to modify the behavior over an upcoming time period.

In some aspects, the techniques described herein relate to a system or a computer-readable medium that perform the above recited techniques.

Implementations of the described techniques may include hardware, a method or process, or computer software on a computer-accessible medium. The details of one or more implementations are set forth in the accompanying drawings and the description below. Other features will be apparent from the description and drawings, and from the claims.

The illustrated embodiments are merely examples and are not intended to limit the disclosure. The schematics are drawn to illustrate features and concepts and are not necessarily drawn to scale.

The above-mentioned aspects, as well as other aspects, features, and advantages of the present technology will now be described in connection with various embodiments. The inclusion of the following embodiments is not intended to limit the disclosure to these embodiments, but rather to enable any person skilled in the art to make and use the contemplated claimed subject matter. Other embodiments may be utilized, and modifications may be made without departing from the spirit or scope of the subject matter presented herein. Aspects of the disclosure, as described and illustrated herein, can be arranged, combined, modified, and designed in a variety of different formulations, all of which are explicitly contemplated and form part of this disclosure.

102 1500 1550 Described herein are systems and methods that utilize large language models and/or generative artificial intelligence to assess and change human behavior associated with managing user health, lifestyle, and diabetes. In some implementations, the systems and methods may utilize one or more devices (e.g., device, bands, or band) for performing non-invasive glucose monitoring, behavior assessment, and recommendation/actionable instruction/insight generation. In some implementations, the systems and methods described herein may monitor, determine, and inform the user about an anticipated or recent glucose reading and provide information about a recent trend in glucose readings for the user. In some embodiments, the systems and methods described herein can perform such monitoring to facilitate obtaining health knowledge for a user and to assist in user adherence to lifestyle improvements to reduce risks for developing blood sugar related illness, including, but not limited to: Type 2 Diabetes (T2D), Type 1 Diabetes (TID), and prediabetes. In some embodiments, the systems and methods described herein may be used to enable early diagnosis of Type 2 Diabetes (T2D), Type 1 Diabetes (TID), and/or prediabetes. In some embodiments, the systems and methods described herein can generate predictive glucose responses and predictive health related events. In some embodiments, the systems and methods described herein can use data obtained from glucose monitoring to facilitate predictions that may be used to generate instructions for a user to carry out to treat, manage, and/or prevent blood sugar related illnesses.

One technical problem sought to be solved by the systems described herein is to provide non-invasive, near real time and/or continuous glucose and/or behavior monitoring for a user. An example technical solution provided by the embodiments described herein includes employing at least one sensor device, software algorithms, and/or one or more machine learning (ML) models to assess glucose changes and/or related behavior over time for a user. For example, the systems described herein may include a device arranged to capture images of a body site including at least one venous structure (e.g., a vein). The systems described herein may include one or more of: a wearable device, a server(s), an artificial intelligence model, or the like. Such systems may analyze the images to determine blood glucose levels and/or blood glucose changes. In some embodiments, the systems described herein may detect/determine blood glucose changes over time and in response, may generate content such as recommendations, actionable instructions, and/or insights for a user wearing or otherwise accessing a wearable device arranged to capture images of a body site having at least one viewable venous structure. In some embodiments, the content may be based on the detected/determined glucose changes for a user. In some embodiments, the content may be based on detected/determined glucose changes for a user and historical glucose data for the user. In some embodiments, the content may be based on detected/determined glucose changes for a user, historical glucose data for the user, and/or historical glucose data for any number of other users. In some embodiments, the content includes educational content selected according to glucose behavior, tracked eating behavior, and/or tracked exercise behavior.

The detected and/or monitored behaviors described herein may be managed, coached, or otherwise modified by the systems described herein. For example, the systems described herein may implement comprehensive behavioral and physiological monitoring across multiple categories to enable personalized health interventions and coaching. The behaviors may include, but are not limited to one or more of: dietary behavior, physical activity behavior, sleep and circadian rhythm behavior, stress or emotional behavior, communication and/or technology usage pattern behaviors, location-based or environmental behaviors, physiological behaviors or measurements, and hormonal and temporal behaviors, etc.

For example, dietary behavior management may include the systems described herein monitoring a frequency and timing of meals and tracking specific meal content. Such a system may analyze tracked eating behavior and can identify high added sugar consumption as a modifiable risk factor including monitoring consumption of sugar-sweetened beverages (e.g., soda, candy, fruit drinks, etc.) and/or refined carbohydrates (e.g., white bread, white rice, pastries, etc.). In some embodiments, the systems described herein may further monitor alcohol intake patterns, including frequency and context (e.g., social functions, weekends) to provide behavioral guidance, and may track eating timing such as monitoring adherence to a cutoff time for evening eating (e.g., 8 or 9 PM), to assist and/or coach the user with ways to minimize late-night glucose spikes.

Physical activity behavior may be monitored by the systems described herein by detecting and/or analyzing exercise patterns including tracked exercise behavior and activity level and frequency including both aerobic and strength-based exercises. The system may use accelerometers and/or gyroscopes to track movement changes and can detect when there is little physical activity sensed, while also tracking post-prandial light exercise and monitoring whether users engage in moderate intensity exercise per day.

Sleep and circadian rhythm events may further be monitoring to assess sleep duration and quality as these represent modifiable risk factors for chronic diseases such as cardiovascular disease and diabetes. The systems may track various sleep events (e.g., excess sleep, sleep deprivation, exhaustion, etc.) and may assess how such events affect (or affected) glucose levels over a time period.

109 102 109 Another technical problem sought to be solved by the systems described herein is to determine which recommendations to provide to a user based on real time glucose measurements and contextual factors such as user location. Conventional systems lack the ability to dynamically adapt recommendations and nudges (e.g., prompts, information, triggers, comments or other text or images that function to generally steer a user to accomplish a goal/task) based on changing glucose levels, behaviors, and/or situational context corresponding to the user. To solve this problem, the systems and methods described herein implement an agent model (e.g., agent model) as a technical solution to the technical problem. The agent model is a computer-using agent that combines one or more AI models to simulate controlling computer interfaces and task performance on or for the wearable device. The agent modelcan enroll trusted members, provide updates and monitoring data, and generate personalized feedback on how past and future diet, exercise, behavior, and medications may affect blood glucose levels.

The technical effect of this technical solution is the provision of personalized, context-aware recommendations and nudges to users as a basis for improving glucose management and overall health. The systems and methods described herein enable real-time adaptation of recommendations based on intermittent or continuous glucose measurements and user location. The systems integrate data from multiple sources to generate more relevant and effective nudges, and improves user adherence to health goals through personalized and timely interventions. The systems enhance the ability for users to plan ahead for patterns (e.g., patterns of glycemia, patterns of eating, etc.) that may negatively impact glucose levels and provides immediate feedback on the impact of lifestyle choices. By using AI-driven, personalized communication strategies, the systems described herein increase user engagement. These technical effects result in a more effective and user-friendly system for managing glucose levels and promoting healthier lifestyle choices, addressing limitations of conventional glucose monitoring and management systems.

In general, the sensor devices described herein may acquire whole or partial images of a target body site, including, but not limited to, venous structures and cutaneous tissues (e.g., skin surfaces). The captured image data may be processed to determine one or more optical characteristics of the venous structures, including a colorimetric value or shade, such as luminance or darkness of the venous structure(s). For example, the systems described herein may be operable to detect and quantify chromatic properties of the skin and vein tissues, including skin tone and vein coloration/saturation. The systems described herein may further analyze morphometric parameters, including but not limited to, a diameter, a cross-sectional area, and/or a volumetric capacity of the vein(s). In some embodiments, the sensor devices may evaluate surface texture attributes of the skin proximate to the veinous structures. In some embodiments, the systems described herein may detect and characterize light interaction phenomena associated with the veins, including backscattering and/or optical absorption.

229 229 229 229 229 Such image analysis may be performed by the systems described herein and may be representative of particular particles, elements, or glucose flowing through the veins. For example, images captured by the image sensors described herein may be analyzed to determine correlations between changes in blood glucose levels by assessing changes in the appearance of the veins in the captured images. For example, the systems described herein may determine a concentration of glucose using a near infrared (NIR) optical sensor, for example, positioned substantially adjacent to a body site of a user. In some embodiments, the systems may determine a concentration of urea by analyzing image data captured by the NIR optical sensor (or another onboard sensor). In some embodiments, the systems may determine a concentration of sodium by analyzing the image data captured by the NIR optical sensor (or another onboard sensor). In some embodiments, the systems may determine a concentration of chloride by analyzing image data captured by the NIR optical sensor (or another onboard sensor). In some implementations, the systems described herein may determine a concentration of potassium using the NIR optical sensor (or another onboard sensor). In some embodiments, the systems described herein may determine a concentration of bicarbonate based on analyzing image data captured by the NIR optical sensor (or another onboard sensor).

The systems and methods described herein may provide the advantages of a wearable solution that reduces or eliminates the use of needle pricks, chemical patch replacements, under skin wire applications, or the in-clinic medical testing for blood sugar assessment. While there are numerous SMBG systems available, the embodiments described herein provide advantages compared to conventional solutions of blood sugar assessment and management.

For example, while conventional methods of non-invasive measures for SMBG have shown that they can provide reliable measurements related to blood glucose levels, such methods, despite these positive results, have been limited due to technical or usability issues with the approaches. For example, conventional methods of non-invasive measures for SMBG may include componentry that may be placed in body regions that may cause the componentry to be rubbed, moved, dislodged or the like. Such componentry may fail when improperly moved or dislodged. In some embodiments, conventional componentry for SMGB may also include electrochemical portions that have to be replaced or refilled.

The embodiments described herein may function to address at least the above limitations of conventional methods of SMBG. For example, the systems and methods described herein may provide a non-invasive and wearable glucose monitor that is an affordable, instantaneous, general wellness device. This device, coupled with a software application can function to provide meaningful and actionable blood sugar and/or general health insights for a particular user wearing the device. In addition, wearing one or more of the devices described herein may promote a healthy lifestyle ultimately helping users reverse prediabetes and/or prevent T2D. The systems and methods described herein provide at least the advantages of zero medical waste, user friendly interfaces, and in general, a low barrier to adoption thereby aiding widespread adoption.

In operation, the systems and methods described herein can illuminate a body site and capture images of the body site using an image sensor, such as an NIR optical image sensor (e.g., a camera) or a camera using a red-light source, to assess near infrared light absorption and scattering. For example, the image sensor may capture images of a body site of a user wearing a device housing the image sensor. The image sensor may capture images of a combined effect of both a physiological change in a human body and a physical light scattering property at the body site. In such examples, the body site may be illuminated with a wavelength of light in the range of highest sensitivity of the NIR sensor, that is, highest quantum efficiency of the NIR sensor and the light source wavelength may range from about 700 nanometers (nm) to about 1200 nm. The captured one or more images may be of any part of the human body. For example, a body site, as used herein may include one or more of a wrist (e.g., volar, dorsal), fingers (e.g., palmar or dorsal), an arm, a knee (e.g., popliteal fossa), a palm, a leg, an ankle, a foot, toes, etc. In operation, the systems and methods described herein may analyze captured images to determine the effects of physiological changes by analyzing how the body site scatters the illuminated light.

While the systems and methods described herein utilize the infrared or the near infrared spectrum to noninvasively assess glucose levels, spectroscopic techniques are not utilized in the techniques described herein. Instead ML models are generated and used to map image content of venous blood vessels to accurate glucose levels.

102 For health and wellness in the prediabetes space, the conventional inputs for lifestyle change continue to be useful. For example, diet intake, activity level and frequency, sleep duration and quality, and stress level are modifiable risk factors for chronic diseases, especially for cardiovascular disease and diabetes. The systems and methods described herein may monitor or retrieve any one or more of: diet intake, activity level and frequency, sleep duration and quality, and stress level of a user wearing the device. The data obtained from such monitoring or retrieval may be used as an input into a generative AI model for causing changes in user behavior. The systems and methods may utilize any number of functions, tools (e.g., food logging tools, sleep monitoring tools, step or physical activity monitoring tools, stress measuring tools, etc.) and AI agentic workflows to generate recommendations for modifying one or more risk factors associated with chronic disease. Mobile device-type measurements may include one or more of: frequency of phone/app interaction and patterns of use, glucose spike patterns, frequency and timing of meals, frequency and pattern of communication via text message, email message, phone call, or voice message, location services (e.g., restaurant, game, work, home, vacation), and social networks of individuals in a same location and an associated behavior pattern.

102 For human behavior, any one or more of the above-recited inputs from a mobile device regarding patterns of communication and behavior, plus data on blood glucose from deviceor another source, location, and low-burden user provided data on age, gender, comorbidities. For example, user behavior such as regularly ingesting a food type (e.g., pizza, ice cream, sugar, etc.) on a particular day and/or performing or not performing exercise on such days may be assessed by the systems described herein to determine a behavior pattern and to generate one or more suggestions, recommendations to allow the user to plan ahead for such patterns to mitigate such patterns or to enable healthier choices after the pattern occurs.

Together, with an abundance of data, the systems and methods described herein may use AI/ML techniques to generate distinct patterns of behavior, and can iterate on these patterns with increased inputs to further understand which patterns are the riskiest.

209 210 216 After an understanding of behavior patterns that have a higher risk of glucose spikes, the systems and methods can use LLM/AI (e.g., generative system, prediction system, and/or NNs) to nudge (e.g., prompt, inform, trigger, comment or steer) the consumer to move away from performing particular behaviors. For example, the systems and methods described herein may warn against drinking more than two drinks if the user is at a bar with friends at night, as this can cause a significant glucose spike. If the user is alone at home during the day and there is little physical activity sensed, LLMs or other AI models and/or agents can nudge the user to move. If being home alone is a context that has previously led to over-eating and high glucose, use of this technology may pre-emptively suggest foods high in fiber and protein instead of the high refined-carbohydrate snacks that are easiest to grab while at home. For example, the systems and methods described herein may utilize an LLM or other AI model or agent to suggest a lighter meal after a prolonged glucose spike after a meal at a restaurant.

Secondly, glucose time in range may drive the deployment of the health behavior-related nudges. The personalized portion of this process may be based on trial and error of specific tactics for each user. If post-prandial glucose is continually highly elevated after meals, the first nudges may be to decrease refined carbohydrates and increase high fiber and protein in each meal. If the user is deployed this suggestion and glucose spikes do not change for 3 days postprandially, it may suggest changing meal order (start with the vegetables and protein-based foods, end with eating carbohydrates). If that does not work, the systems and methods described herein may suggest post-prandial light exercise and nudge towards 30 min moderate intensity exercise per day. If the user gets bored of the same type of exercise daily, the systems and methods described herein may suggest a more vigorous HIIT related workout at times of the day when there is little activity, and while are unscheduled based on a calendar associated with the user.

228 229 The format of the nudges may vary: short games, popups, text messages, alerts based on blood glucose levels, prompts for logging diet intake, etc. In some embodiments, the nudges may provide an advantage of increasing user involvement and enhancing a sense of accomplishment for the user. The systems described herein may use inputs from sensors,, geolocation, user-entered data, etc. to define patterns of risk and try to sway the individual away from these risky behaviors. In some implementations, the devices and methods described herein may operate in response to user provided permissions to ensure user privacy, HIPPA privacy compliance, and the like.

In some implementations, the systems and methods described herein may employ motivational interviewing. Motivational interviewing may assist the systems described herein in capturing user-specific barriers and may direct the user towards a particular behavior change. This can be an iterative process to detect a behavior change and incrementally improve the behavior based on recommendations and/or nudges. In such examples, the systems described herein can conduct the motivational interview and generate the steps to trigger behavioral changes based on the user answers in the interview and/or prior or ongoing monitoring of the user activity or behaviors.

1 FIG. 100 100 100 228 102 102 228 102 102 illustrates an example high level systemfor obtaining noninvasive glucose assessments for a body site of a user and monitoring well-being and/or glucose and generating recommendations or other content for behavioral and/or lifestyle changes. The systemmay use images and ML techniques to detect and analyze glucose signals associated with a body site of a user to determine whether the signals are coming from interstitial fluid or blood. For example, the systemmay use one or more noninvasive optical sensors (e.g., image sensor(s)) on a wearable deviceto detect physical and physiological changes in venous blood under a skin portion of a user. In some embodiments, the wearable device(and coupled image sensor(s)) may be positioned against (or near to or substantially near to) the body site. In some embodiments, the wearable devicemay be worn on a body part and the body site may be placed in contact (or in near contact or in substantially near contact) to the wearable device.

The sensor devices described herein may include at least one image sensor (e.g., optical sensor, image capture device, camera, etc.) that may capture images of visual changes that occur in one or more veins (e.g., venous blood vessels) near to or at a surface of the body site of the user. For example, the sensors described herein may capture/observe a combined effect of both osmolality changes in an appearance of one or more veins of the body site and changes in a light absorption property of water and glucose associated with the one or more veins at the body site. These appearance changes may be used by the systems described herein to determine when a glucose level has changed for the user.

1400 102 110 14 FIG. 1 FIG. 2 FIG. In some embodiments, the systems and methods described herein may provide a non-invasive optical sensor that detects a combined effect of both a physiological change in the human body and the physical light scattering property. Such changes/effects may be analyzed using a first machine learning (ML) model and/or a first Deep Learning (DL) model that may determine if the physical and/or physiological changes are coming from blood in the veins, for example. A second ML model (which may represent a DL model) may map the changes in blood to a glucose level. A software application executing on a computer system (e.g., computer systemofand/or wearable deviceof/) may generate meaningful and actionable insights and/or recommendations for a user wearing the optical sensor and/or may provide information, insights, and/or recommendations to a health care provider, a caretaker, a buddy device, or the like.

102 102 104 104 102 104 104 102 106 102 110 102 110 104 106 The wearable devicemay include any number of image capture devices and/or sensors to capture images of the body site. The devicemay utilize one or more ML modelsto discern any physiological changes in venous structures and/or skin in the images to ascertain a glucose level or glucose change in the blood associated with the body site versus a physiological change in glucose in the interstitial fluid associated with the body site. In some embodiments, the ML modelsmay map the detected changes in the blood to particular glucose levels. The wearable devicemay obtain additional glucose signals over time by monitoring the body site. Data may be generated from the monitoring, and the ML modelsmay analyze changes over time. Such changes and data may be detected by the ML modelsusing one or more processors as described herein. The changes and/or data may be used to generate user interface content and/or data for an application that may be presented on an optional display of the wearable device. In some embodiments, the user interface content and/or data may be presented on a computing devicethat is in wired or wireless communication with the wearable deviceor a buddy devicein communication with the wearable device. The buddy devicemay represent a device of a trusted person that a user of wearable devicechooses to share health data or other related device data. The computing devicemay be a smart phone, a smart watch, a mobile gaming device, a laptop computer, a desktop computer, a cloud hosted computer, or the like.

110 110 109 102 110 109 102 102 109 102 109 102 110 110 The buddy devicemay be a smart phone, a smart watch, a mobile gaming device, a laptop computer, a desktop computer, a cloud hosted computer, or the like. The buddy devicemay receive output from the agent modeland provide insights, recommendations or the like to the user of the wearable device. The output may include progress notes, blood glucose or other health changes, diet selections, and the like. A buddy devicemay access the agent modelas a way to enroll in receiving input from the wearable deviceaccording to predefined user permissions. In some embodiments, the wearable deviceincludes the agent modeland automatically generates the output based on a user-configured schedule selected by a user of the wearable device. In some embodiments, the agent modelis external to deviceand may be triggered externally by the buddy deviceaccording to a buddy-configured schedule selected by a user of the buddy device.

109 109 209 210 109 102 209 210 109 100 200 100 200 102 108 b b In some embodiments, the agent modelmay represent a computer-using agent that can perform tasks on behalf of a user or device. For example, agent model agent may be a computer-using agent model that combines one or more other AI models to simulate controlling computer interfaces and/or task performance. In some embodiments, the agent modelmay use or employ generative systemand/or prediction system. In some embodiments, the agent modelmay be outside of deviceand may employ generative systemand/or prediction system,. In some embodiments, the agent modelmay execute on a separate computing device outside of systems,, but may be communicatively coupled to system,to provide input to wearable deviceand/or lifestyle AI coach.

104 104 209 209 210 210 104 104 104 104 200 a b a a a The ML models,described herein may function in conjunction with additional AI components including generative systems,and prediction systems,where the generative systems may function with any or all ML models,to assess user behavior and user indications. The prediction systems may function with ML models,to preprocess, optionally segment, analyze, and predict blood glucose levels. This ML architecture enables the systemto process complex physiological data and generate accurate predictions for personalized health recommendations and behavioral interventions.

109 109 109 102 110 109 102 110 109 102 109 102 109 108 2 FIG. 1 FIG. In some embodiments, the agent modelmay operate in a continuous loop to send instructions to be executed on a computer or browser (e.g., app). Example instructions for the agent modelmay include requests to view updates on behavior, food intake, exercise, glucose level changes, etc. In some embodiments, the agent modelmay capture screenshots of such data from the wearable deviceand provide the screenshots as updates to buddy device. In some embodiments, the agent modelcan take additional actions including, but not limited to triggering glucose monitoring for a time period, triggering comments, insights, and/or recommendations to be provided to the user of device, and/or buddy device. While the agent modelinis depicted within wearable device, the agent modelmay instead be external to wearable device, as shown in. In some embodiments, the agent modelmay be part of lifestyle AI coach.

102 102 108 108 102 Some or all components of the wearable devicemay be integrated into a band, a watch, a smart watch, an adhesive strip, jewelry, a patch, clothing, or any other wearable accessory. For example, some or all components of the wearable devicemay be incorporated into a band or a watch or a pair of bands or pair of watches to be worn one on each arm or leg (or other appendage on the arm or leg). In some embodiments, the lifestyle AI coachmay operate as a server, cloud server, or other hosted computing device. In some embodiments, coachmay operate locally on wearable deviceand may provide an interface in which to ask questions or otherwise inquire about future goals, shopping or food based data, exercise data, monitored data, and/or wearable device captured data. In response, the interface may generate and provide content responsive to such inquires.

200 102 102 In operation, the systemmay analyze images to determine and/or monitor blood glucose changes for a user. The images may be captured by wearable deviceand may depict a body site of the user. For example, if the wearable deviceis a band, the body site may be a portion of a wrist, arm, ankle, leg when the band is worn on the body. The body site may include skin and a visual full or partial view of one or more venous structures.

104 109 102 In some embodiments, the ML modelsand/or agent, in general, may utilize glucose assessments and/or user input to determine content such as recommendations, insights, and/or actionable guidance including, but not limited to maps, directions, shopping lists, educational content, menu items, or the like. The content may be provided to the wearable deviceto give the user personalized feedback on how past and future diet, exercise, behavior, and/or medications may affect blood glucose levels so that the user may use the information to adjust lifestyle choices.

200 In some embodiments, such content may be generated as a basis for coaching the user to improve health metrics and/or lifestyle metrics. For example, the content may include lifestyle change recommendations that may be used as a way to lower a burden of healthful eating for a user to move the user behavior from less healthy states to more healthy states. For example, content generated by systemmay include food recommendations and/or timing of ingesting such recommended foods. In some embodiments, recipe recommendations may be generated and provided to the user. In some embodiments, grocery store recommendations may be generated and provided to the user. In some embodiments, grocery purchase recommendations may be generated and provided to the user. In some embodiments, goods or product recommendations may be generated and provided to the user. In some embodiments, educational recommendations may be generated and provided to the user. In some embodiments, educational content recommendations (e.g., advice, links to research, details about foods, etc.) may be generated and provided to the user along with any other recommendation described herein.

200 200 200 200 102 108 102 In some embodiments, the systemmay detect/determine blood glucose changes over time by analyzing the images of the body site. In response, the analysis of the changes, the systemmay generate content, as described herein, based on the detected/determined glucose changes. In some embodiments, the content may be generated by systembased on user specified requests or input. In some embodiments, the systemmay generate content according to data from wearable deviceand/or direct requests to lifestyle coachfrom the user using device.

200 108 108 219 102 102 In some embodiments, the systemmay include or be communicatively coupled to a lifestyle coach. The lifestyle coachmay include an AI chat agent such as a large language model (LLM)that may provide content that takes the form of coaching advice, lifestyle advice, recommendations, educational content, or the like based on data captured by wearable deviceand/or data requested by a user of wearable device. In some embodiments, the content may include shopping recommendations, restaurant recommendations and/or corresponding educational content, as described elsewhere herein.

200 110 110 102 110 102 102 110 252 102 110 The systemmay include or be communicatively coupled to one or more buddy devices. A buddy devicemay represent a device of a trusted person that a user of wearable devicechooses to share health data or other related device data. The buddy devicemay receive data from devicewith permission of the user of device. A user of buddy devicemay review the data and provide comments and/or recommendationsto encourage the user of deviceto improve lifestyle choices and/or behaviors. In some embodiments, the buddy devicemay function as a motivational check in and/or other process/entity capable of motivating the user to continue on an improved health path.

2 FIG. 200 200 100 200 102 108 110 102 108 102 252 110 illustrates a block diagram of an example systemfor monitoring well-being and/or glucose and generating recommendations or other content for behavioral and/or lifestyle changes. In some embodiments, the systemmay represent a detailed version of system. As shown, systemincludes wearable devicecommunicatively coupled to a lifestyle AI coachand a buddy device. The wearable devicemay interact with and/or receive recommendations and coaching from coach. The wearable devicemay interact with and/or receive recommendationsand input from one or more buddy devices.

200 108 108 219 102 102 In some embodiments, the systemmay include or be communicatively coupled to a lifestyle coach. The lifestyle coachmay include an AI chat agent such as a large language model (LLM)that may provide content that takes the form of coaching advice, lifestyle advice, recommendations, educational content, or the like based on data captured by wearable deviceand/or data requested by a user of wearable device. In some embodiments, the content may include shopping recommendations, restaurant recommendations and/or corresponding educational content, as described elsewhere herein.

108 102 108 102 108 102 200 The lifestyle coachmay function to provide feedback on a meal just consumed by a user. The feedback may be based on detected or determined glucose responses captured by wearable device. In some embodiments, the lifestyle coachmay suggest content and/or timing of a next meal with a goal to minimize glucose spikes in the blood glucose level of the user using device. In some embodiments, the lifestyle coachmay provide feedback that includes motivating and/or incentivizing content for enabling a user to avoid glucose spikes for a set time period. The suggested time period may be based at least in part on a glucose history captured for the user by device, for example. The set time period may be predefined, user defined, and/or systemdefined and may be set for about 1 day to about 10 days; about 1 day to about 2 days; about 1 day to about 3 days; about 1 day to about 4 days; about 1 day to about 5 days; about 1 day to about 6 days; about 1 day to about 7 days; about 2 days to about 4 days; about 3 days to about 4 days; about 3 days to about 5 days; about 4 days to about 5 days; about 4 days to about 6 days; about 5 days to about 6 days; about 6 days to about 7 days; about 7 days to about 8 days; about 8 days to about 9 days; or about 9 days to about 10 days.

108 102 108 In some embodiments, the lifestyle coachmay access databases of historical data for the user of wearable device. In some embodiments, the lifestyle coachmay access databases of historical data for prediabetic or diabetic users. Example databases may include, but are not limited to open-access glycemic data, which may include one or more datasets that contain glucose values from subjects with prediabetes or diabetes. The data may represent a set time period (e.g., data collected over 2-4 weeks at regular intervals of 5 minutes, with diet intake and mealtimes documented). Other user-specific or user-agnostic databases may be utilized. The databases accessed may include glucose values collected for shorter or longer time periods. The databases may represent glucose value captures performed intermittently or with a continuous glucose monitor.

108 The historical data may be used to generate recommendations, content, or other insight. For example, the lifestyle coachmay be programmed to perform intermittent checks with instructions such as ‘If the user is found to have high postprandial glycemia of 140-199 mg/dL two hours after a meal, recommend the user to increase protein and healthy fat intake in the next meal, and eat these food components before a small portion of complex carbohydrates. Keep this recommendation in your non-parametric memory when suggesting menu items from the user-provided restaurant or while recommending grocery items. Ask the user to set reward goals at the beginning of each week. Some example reward goals are a dessert with suspension of the measurement of post-prandial glycemia akin to a cheat dessert after 3 days of glycemic control or a manicure after 5 days of controlled glycemia. Nudge the user to achieve these rewards with progressive, adaptable goals for glycemic control.’ Other instructions are of course possible, as one of skill in the art will appreciate.

200 200 200 200 200 In some embodiments, the blood glucose data may be obtained from a known database of multiple users or from a historical record of data of a particular user. The systemmay be prompted/triggered to perform particular tasks based on received, monitored, or otherwise detected blood glucose. For example, if a user is found to have high postprandial glycemia of about 140-199 mg/dL about two hours after a meal, the systemmay recommend the user to increase protein and healthy fat intake in the next meal, and eat these food components before a small portion of complex carbohydrates. The systemmay also be prompted or programmed to keep these types of recommendations in non-parametric memory when suggesting menu items from a user-provided restaurant or while recommending grocery items, for example. The systemmay also be prompted or programmed to ask the user to set reward goals at the beginning of each week. Some example reward goals include a dessert with suspension of the measurement of post-prandial glycemia akin to a cheat dessert after 3 days of glycemic control or a manicure after 5 days of controlled glycemia, for example. The systemmay be further prompted or programmed to nudge the user to achieve these rewards with progressive, adaptable goals for glycemic control.

200 108 200 In some embodiments, the systemimplements a lifestyle AI coach (e.g., lifestyle AI coach) that generates and provides personalized health recommendations, with a focus on glucose management. To do so, the systemmay utilize one or more Application Programming Interface (API) keys to import libraries, databases, and/or functions to enable personalized health assessments and recommendation generation. These may include APIs for accessing external services such as mapping and restaurant information databases, web scraping, online shopping platforms, etc.

200 200 200 200 At a high level, the systemmay define and utilize a set of software functions for interacting with external services. For example, a get_grocery_stores function may be used to find nearby grocery stores based on a provided ZIP code. In some embodiments, the systemmay define and utilize a sum_item_prices function may be used to adjust a list of items to fit within a specified budget. In some embodiments, a get_restaurant_dishes function may be used to retrieve menu items or other purchasable items for a specified restaurant or other venue. In some embodiments, the systemmay define and utilize a check_restaurant_menu_availability tool to check if a particular restaurant menu or storefront is available to access through system. The tool may function when the user is dining out to analyze a restaurant menu in real time, and suggest modifications to dishes that could help maintain stable glucose levels.

108 200 202 In some embodiments, the lifestyle AI coachis a conversation agent that utilizes one or more language models. This agent may initialize with a systemprompt that defines the coach role and capabilities. In addition, the agent may use memoryto maintain conversation history. Further, the agent may implement a conversation chain that processes user input and generates responses.

200 204 17 18 FIGS.A-D In some embodiments, the systemmay implement and/or generate a user interface (e.g., shown in. The user interface may be a web interface or an app interface (e.g., apps). The interface may display logos, application titles, menus, conversation history containers, input fields, and response fields.

108 108 In some embodiments, the lifestyle AI coachmay recommend nearby grocery stores and healthy food items based on a determined user location. The coachmay suggest healthy menu options from restaurants, provide personalized health advice based on glucose levels and user behavior, adapt recommendations based on user location and context, and/or generate nudges or messages to encourage healthy behaviors.

200 108 102 102 102 200 The systemintegrates various data sources to provide comprehensive recommendations responsive to user input received through the lifestyle AI coach interfaceand/or directly through the wearable device. The data sources may include glucose measurements, which may be simulated or determined and sent by wearable deviceor another wearable device coupled to device. The data sources may further include location data to enable provision of context-aware recommendations by system. The data sources may further include restaurant and/or grocery store information obtained from external databases and/or scraped online sources.

200 The systememploys artificial intelligence techniques to analyze patterns in user data and generate personalized recommendations. This may involve machine learning models trained on health and nutrition data, glucose data, and/or user-specific health data.

200 110 110 102 110 102 109 102 In some embodiments, the systemmay include or be communicatively coupled to one or more buddy devices. A buddy devicemay represent a device of a trusted person that a user of wearable devicechooses to share health data or other related data. The buddy devicemay have access to data provided by wearable devicethrough agent model. Access is provided according to user selected privacy conditions configured by a user of wearable device.

109 102 102 109 109 102 109 102 102 The agent modelmay represent an agent that can control portions of data from wearable deviceand/or particular actions performed by (or associated with) wearable device. For example, the agent modelmay be a computer-using agent that can perform tasks on behalf of a user or device. In some embodiments, the agent modelmay be a computer-using agent model that combines one or more other AI models to simulate controlling computer interfaces and/or task performance of wearable device. The agent modelmay function to enroll a trusted member (e.g., user, user device, etc.) and may cause deviceto provide updates and/or monitoring data such that the trusted member may use such updates and/or data to provide accountability for goals of the user associated with wearable device.

102 110 102 109 110 109 110 109 102 102 109 110 102 110 102 109 102 109 110 102 110 109 109 110 109 110 In a non-limiting example, the agent model may function as an automated agent to obtain progress data, monitoring data, or the like from wearable device. Such data may include obtaining glucose levels for a user, summarizing the data, and sending such glucose data to the buddy device. In some embodiments, the data is not summarized, but instead sent as a screenshot, value, or interface depicting a portion of monitored data pertaining to the user of device. The agent modelmay receive feedback from a user of the buddy devicethrough an interface that communicatively couples the agent modelto the buddy device. The agent modelfunctions to operate as an agent for the wearable device(e.g., a band) and/or as an agent for a device (e.g., smart watch, phone, etc.) communicatively coupled to device. The agent modelcan send data as messages from the buddy deviceto the wearable deviceand may receive responses from the buddy deviceand the wearable device. In some embodiments, the agent modelmay include or have access to databases, data, web scrapes, or the like in order to provide content, updates, or other requested information to the wearable device user and/or the buddy device user. In this example, the user of the wearable devicemay provide permission to the agent modelto share updates and/or health data with a selected one or more buddy devices. Future updates and/or data sharing from the deviceto the one or more buddy devicesmay be automatically performed by the agent modelaccording to user enabled permissions. In some embodiments, the agent modelmay be used by a user of the buddy deviceto offer feedback or encouragement to the user based on what the agent modelprovided to the buddy deviceas an update.

109 102 109 109 109 In some embodiments, the agent modelmay function to find and provide options to the user of wearable device. For example, the agent modelmay analyze a current location for the user and glucose trends corresponding to the user to suggest nearby restaurants with menu items suitable for maintaining stable blood glucose levels. If the user is at work, the agent modelmay recommend healthier alternatives to common vending machine snacks. When grocery shopping, the agent modelmay provide real-time suggestions for lower glycemic index alternatives as the user scans or selects items.

109 109 109 In some embodiments, the agent modelmay function to provide an automated end-to-end ordering process for food, groceries, supplies, or other purchasable items. For example, if the glucose levels for the user are detected to indicate that the user may benefit from more fiber, the agent modelmay generate a shopping list of high-fiber foods, locate these items at nearby grocery stores, compare prices, place items in an online shopping cart and/or place an order for delivery or pickup. Similarly, if the user frequently skips meals due to a busy schedule, the agent modelmay suggest and order healthy meal kit deliveries that align with the glucose management goals corresponding to the user.

109 109 109 109 109 In some embodiments, the agent modelmay provide real-time updates on ingredients, menu items, suggestions, grocery store inventory, etc. For example, if a user is dining out, the agent modelmay analyze the restaurant menu in real-time, suggesting modifications to dishes that could help maintain stable glucose levels. The agent modelmay also alert the user to new low-glycemic products available at a predefined or selected grocery store. During seasonal changes, the agent modelmay update recommendations to include seasonal produce that can positively impact glucose management. If a user is determined to have a favorite healthy item and that item is out of stock, the agent modelmay immediately suggest alternatives with similar nutritional profiles.

108 110 109 110 109 In some embodiments, the lifestyle coachmay ask the user for contact information for a buddy device(e.g., email, phone number, etc.). In some embodiments, the agent modelmay suggest particular users in a contact list of a wearable device user. The suggested users may be suggested to function as a buddy device. A messaging agent may share the same guidelines and incentives, proposed by the LLM, with the trusted member. The message may help the buddy understand how particular support may help the user. The messaging agent (e.g., agent model) may also share progress associated with the user. Using the function calling ability of the LLM, the LLM may call a web-searching API to find places/stores that would be of interest to the trusted member that is also in the vicinity of the recommended grocery store, if that trusted member is responsible for groceries and/or cooking.

2 FIG. 102 202 102 202 204 206 208 209 209 210 210 212 104 104 1 104 104 2 216 216 218 220 222 224 226 202 204 206 208 209 209 210 210 212 104 216 104 102 200 104 104 108 102 108 b b a n b n b b b a a b As shown in, the wearable deviceincludes memoryfor storing program instructions and data that may be used and/or generated by processes carried out by the wearable device. The memorymay store and execute instructions associated with applications (e.g., apps), patient data, mapping logic, generative systemand/or generative system, prediction systemand/or prediction system, image analyzer, ML models,and/or ML models,, and/or neural networks,. The processors, the analog to digital converter, the digital to analog converter, the optional antenna, and the optional displaymay also interface with memoryand/or apps, patient data, mapping logic, generative systems,, prediction systems,, image analyzer, ML models, and/or neural networks, as described in detail below. While the ML modelsare depicted as executing on the wearable device, other implementations may provide ML models external to system. In some embodiments, the ML modelsmay be replaced with external models, which may operate in lifestyle AI coach(or another computing device communicatively couplable to deviceor coach).

102 102 228 102 229 102 229 102 102 228 The wearable devicemay include one or more sensors for detecting physiological parameters of the user. For example, the wearable devicemay include one or more image sensorsfor obtaining visually-identified glucose measurements, for example. In some embodiments, the wearable devicemay include one or more sensorsthat include inertial measurement units such as accelerometers and/or gyroscopes to track movement changes, temperature sensors to track user temperatures. Other sensors and/or elements may be incorporated into or near to the device. For example, other optical, electrical, and/or mechanical sensorsmay also be included in deviceto measure any number of physiological parameters of a user wearing device. In some embodiments, the image sensorsmay include one or more of: a light source, a red-light source, an IR light source, or a near infrared (NIR) light source for illuminating a body site.

102 228 228 228 102 102 In operation of deviceusing sensor(s), incident light undergoes reflection, absorption, and scattering that may be captured by the image sensor(s). For example, the reflected light arrives at the image sensor(e.g., a camera lens). The camera lens focuses the reflected light onto the sensor which then produces an image that is captured using software described herein. The digital image is stored on a persistent storage medium for later use by components of wearable deviceand/or other computing device(s) communicatively coupled to device.

228 102 102 228 102 228 232 232 228 232 228 102 102 224 102 102 102 102 102 In some embodiments, glucose measurements may be obtained using an optical sensor (e.g., one or more image sensors) that is integrated into or associated with the wearable device. For example, the devicemay include one or more image sensors for detecting a combined effect of osmolality changes in one or more veins of the user by assessing an NIR light absorption property of water and a scattering property of glucose in the veins. In some embodiments, the image sensormay be an NIR camera that captures images of veins in a body site of the user wearing device. In such an example, the image sensormay operate with an optional light sourcewith a wavelength that ranges from about 700 nm to about 1000 nm. In some embodiments, the light source may be a red-light source with a wavelength of about 620 nm to about 750 nm. In some embodiments, the optional light sourcemay be integrated into the image sensor. In some embodiments, the optional light sourcemay be separate from image sensors, but may be installed within wearable device. In some embodiments, the devicemay be communicatively coupled (e.g., via antenna) to an external sensor (not shown) that is not housed in or integrated into device. In such examples, the external sensor may provide data (e.g., images, measurements, etc.) to wearable deviceto determine particular outputs, such as glucose levels, glucose predictions or related predictions, and/or other measurements obtained at a body site of the user wearing deviceand/or the external sensor. In such examples, the external sensor may be part of a watch face while devicerepresents a watch band. The sensors onboard devicemay interface with and otherwise communicate with external sensors onboard a peripherally connected or wirelessly connected other device.

228 102 230 230 228 229 202 218 228 228 229 102 228 In some embodiments, the sensorsand/or other sensors in communication with wearable devicemay be coupled to a sensor interface. The sensor interfacemay store instructions to carry out operations pertaining to received sensor signals from one or more of the sensors,. For example, the instructions may enable interaction with memory, one or more processors, and/or sensorsto communicate sensor data from one or more of the sensors,to the wearable device. The sensor data may be obtained from the sensorsin the form of recordings, measurements, images, and/or audial information.

2 FIG. 104 102 104 209 210 216 104 104 104 104 1 102 102 102 104 104 2 108 104 104 2 209 210 216 104 108 104 104 2 a a a a a a a n b n b n b b b b b n Referring again to, one or more ML modelsmay be optionally installed on wearable device. The one or more ML modelsmay include a generative system, a prediction system, and one or more neural networks. The ML modelsmay include any number of ML models. A first ML model(e.g., ML model) may perform a first set of functions and a second ML model (e.g., ML model) may perform a second set of functions. In some embodiments, the wearable devicemay utilize ML models on a server and/or other device external to wearable device. For example, the wearable devicemay utilize ML modeland/or ML modelon lifestyle coach. The ML models,may include a generative system, a prediction system, and one or more neural networks. The ML models, etc. on coachmay include or provide access to any number of ML models. A first ML model (e.g., ML model) may perform a first set of functions and a second ML model (e.g., ML model) may perform a second set of functions.

104 104 104 104 1 104 2 216 216 216 216 216 a b n n a b a b In general, the ML models(e.g.,,,,, etc.) described herein may include any number of neural networks(e.g., neural networks,, etc.) that are programmed to learn to determine levels of glucose in a vein, for example, by analyzing images of example veins (e.g., training data) at different levels of abstraction in order to determine levels of glucose in images of veins provided to the ML model after the training of the model, for example. In general, the neural networks,, etc. can model complex non-linear relationships amongst inputs to learn how to analyze new inputs. The neural networks described herein may include or use one or more machine learning algorithms to learn from training data and apply that learning to new inputs. The learning may take the form of modeling high-level abstractions in data.

104 104 206 104 102 104 In the examples described herein, the ML modelsmay be trained with training data including images of one or more veins of a user and with data representing glucose measurements, sodium measurements, and/or potassium measurements associated with blood of the respective user. For example, the ML modelsdescribed herein may be trained using (1) a collection of labeled data from clinical study participants to analyze a relationship between glucose, sodium, and/or potassium measurements associated with such study participants, (2) lab generated data values from a basic metabolic panel (BMP) testing and/or analysis. In some embodiments, an example clinical study may include at least about 20 to about 60 participants that are provided a controlled meal prior to a blood draw. Then, blood from a vein for each participant may be sampled (e.g., drawn at regular intervals of about 10 to about 15 minutes for about 3 hours). These blood samples can be run through a lab for BMP testing and/or analysis to obtain a reference value of each participant's blood glucose, sodium, potassium, and blood urea nitrogen (BUN), or other patient data, for example. The ML modelsmay be trained and tested to output a glucose value (or range) for a user wearing the wearable device, for example, based at least in part on elements (1) and (2) above. Although a single clinical study is described, any number of clinical studies and/or any amount of simulated data may be used to train and/or test the ML models.

In some embodiments, the training may include augmentation of datasets used for training. For example, training data may include preprocessing of data used for training. Preprocessing may include, but is not limited to resizing, rescaling, sampling, cropping, rotating, flipping, de-noising, etc. one or more of the images in the training data. In some embodiments, training data may include synthesized new data based on plausible and tested real-life scenarios for a particular user or a population.

104 104 104 104 104 102 a b In a non-limiting example, the ML modelsmay include a first ML model(e.g., ML modelor, etc.) arranged to perform semantic segmentation on images to detect veins (e.g., portions of venous blood vessels). The first ML modelmay perform image segmentation to partition a digital image into multiple segments. This segmenting process may simplify and/or modify a representation of the digital image as a basis in which to locate particular objects, boundaries, or the like within the images. For example, the segmenting process may be performed to identify one or more venous blood vessels and/or venous blood vessel portions within a body site adjacent to the wearable device.

228 In general, the segmenting process may include dividing up one or more digital images (e.g., representing a capture of IR reflections by sensor) and analyzing the one or more digital images. The analyzing may result in identifying particular pixels/voxels in the images and/or applying labels, indicators, markers, or other indications on particular pixels/voxels. In some embodiments, identifying and/or labeling pixels/voxels may pertain to venous blood vessels and/or particular changes in an appearance of such blood vessels. In some embodiments, identifying and/or labeling particular pixels/voxels may pertain to particular changes in light scattering behavior detected from the blood vessels. In some embodiments, identifying and/or labeling particular pixels/voxels may pertain to particular changes in osmolality changes associated with the digital images of the blood vessels. In some embodiments, identifying and/or labeling particular pixels/voxels may pertain to detecting a reduced scattering coefficient detected in the digital images of the blood vessels.

104 102 In some embodiments, segmenting and labeling the digital images may allow that particular features or characteristics in the segmented images be mappable back to features in a corresponding captured image in which the segmented images were derived. In some embodiments, the segmenting process may be used to train the first ML modelto segment structural features of blood vessels captured in images by wearable device.

104 104 104 104 206 The first ML modelmay be one of several ML models within the ML models. The first ML modelmay include a suitable convolutional neural network (CNN) architecture such as a UNET, EFFICIENTNET, RESNET, FR-UNET, Swin-UNETR, OCTA-NET, V-Net, DeepLab, FCN, etc. It is noted, selection of an ML model may self-select the hardware upon which the ML model may be executed. For example, ML models that utilize greater computational resources may not be executed on a mobile device (i.e., the mobile device does not have enough processing power) and instead may be executed on a server and/or desktop. However, it is contemplated that mobile device hardware may improve to become suitable for those ML models that utilize greater computational resources. The first ML modelmay be trained to segment venous blood vessel images and may be understood to include a collection of weights and biases for neurons established during training and details of hyperparameters such as the learning rate associated with the ML model. In some embodiments, patient datamay be used as training data in the event that the user (i.e., the patient under test) has approved the use of such data.

104 104 1 104 104 104 206 n b A second model(e.g., ML modelor ML model, etc.) may receive the output (e.g., segmented images of one or more veins) of the first model as input and may perform a classification of the images. For example, a second model in the ML modelsmay include a suitable architecture, e.g., CNN (e.g., such as one or more deep learning networks), logistic regression, decision trees, random forests, support vector machines, Naive Bayes, K-nearest neighbor (KNN), etc. that may be used for classification of the segmented images into classes of glucose measurements having a range of about 5 mg/dL. The second ML modelmay be trained to assess venous blood vessel images for blood glucose, sodium, bicarbonate, and/or potassium and may be understood to include a collection of weights and biases for neurons established during training and details of hyperparameters such as the learning rate associated with the ML model. In some embodiments, patient datamay be used as training data in the event that the user (i.e., the patient under test) has approved permission of the use of such data.

102 104 In some embodiments, the classification may be based on particular classes (e.g., categories) of glycemia associated with the glycemic index of food. In one non-limiting example, a size of each class may be chosen to be about 5 mg/dL per class. However, the total number of output classes utilized by deviceand ML modelsmay be selected based on the size of each class. For example, the total number of output classes may be determined based on a clinically significant glucose range of about 54 mg/dL to about 250 mg/dL and a range of glucose values obtained from training data. In some embodiments, the number of output classes may be based on predetermined glucose ranges that are clinically significant. In some embodiments, the classification of the segmented images may be classified by the first ML model into classes selected based on one or more clinical trials. Such trials may collect (from each participant in a particular trial) images of venous blood vessels and corresponding blood samples at a same time window and/or before/after the time window. The blood samples may be tested for glucose, potassium, sodium, and/or BUN by a laboratory. Features in the images may be compared to levels of glucose, potassium, sodium, and/or BUN in the blood. Assessments may be made to generate the categories of glycemia. In some embodiments, the categories may be generated and subcategorized according one or more of subject age, disease status, weight, medication status, skin tone, skin thickness, gender, ethnicity, temperature, tracked food intake, stressor(s)/stress level(s), etc.

104 104 228 Upon completion of classification, the training of ML modelsmay be considered completed such that the ML modelscan be used to infer a range of glucose values (e.g., mg/dL), given a sensor output from one or more sensors, for example.

104 104 206 102 In some embodiments, the ML modelsmay be expanded and trained to predict a range of glucose values according to a panel of metabolic readings, including, but not limited to potassium, sodium, and/or BUN, to provide an overall representation of current metabolic status related to prediabetes, for example. In some embodiments, the ML modelsmay be programmed to determine a maximum achievable accuracy, sensitivity, and/or specificity for glucose, potassium, sodium, and/or BUN values. Other patient datamay also be used for particular calculations and/or related predictions generated by devicewhen the user (i.e., the patient under test) has provided permission for the use of such data.

104 104 104 104 216 216 104 a b Example ML modelsmay include a CNN architecture that executes using two or more graphical processing units. For example, a CNN architecture like EFFICIENTNET or RESNET model may be used. In some embodiments, pre-training one or more of the ML modelswith IMAGENET may be performed to assess and/or improve the accuracy of the output of one or more of the ML models. Bias, variance, and error analysis may be performed, on the development set, and appropriate model hyperparameters can be fine-tuned to iteratively increase the performance of the ML models. Further, different layers of the neural networks,can be introspected to understand what the different hidden units in both the shallow and deep layers of the networks are learning or computing. Information from neural network introspection may also be used to improve the performance of one or more of the ML models. Data visualization may be performed using suitable dimensionality reduction techniques, e.g., SNE, t-SNE, UMAP, etc, for example, to understand how the output data is clustered and whether there are gaps in the output data.

204 In some embodiments, the performance of the ML model may be assessed (or results may be displayed) by an appor server/cloud device (not shown). The assessment may evaluate according to any or all of the following elements: sensitivity, specificity, true/false positive, and true/false negative. The sensitivity element may refer to a probability of a positive test, conditioned on truly having a condition such as prediabetes, diabetes, etc. (or tested positive by a predefined test level if the true condition cannot be known). The specificity element may refer to a probability of a negative test, provided one does not have the condition. The true positive element may represent a test result that correctly indicates the presence of a condition or characteristic. The true negative element may represent a test result that correctly indicates the absence of a condition or characteristic. The false positive element represents a test result which wrongly indicates that a particular condition or attribute is present. The false negative element may represent a test result which wrongly indicates that a particular condition or attribute is absent.

226 102 102 204 204 204 228 1400 204 102 102 204 228 102 204 102 102 In some embodiments, one or more user interfaces may be provided on the optional displayof wearable deviceand/or another computing device coupled to wearable device. The user interfaces may be textual, tactile, audial, and/or graphical and may be displayed in one or more apps. The appsmay provide information (via the user interfaces) to support user self-monitoring with the intention of influencing health decisions to prevent the development of prediabetes and/or T2D. In some embodiments, one or more appsmay trigger for the sensorto capture one or more images. The one or more images may then be sent to an external computing device (e.g., cloud server, mobile phone, computing system, or the like) for further assessment and machine learning-based predictions. Any predicted results (e.g., glucose results, potassium results, sodium results, etc.) may be provided back to the app, which may cause display of such results on wearable device(or another computing device external to device). In some embodiments, one or more appsmay trigger for the sensorto capture one or more images. The one or more images may then be analyzed by wearable deviceto perform assessment and machine learning-based predictions. Any predicted results (e.g., glucose results, potassium results, sodium results, etc.) may be provided via the app, which may cause display of such results on wearable device(or another computing device external to device).

204 102 204 204 204 In some embodiments, an app(s)may provide graphical content showing a current glucose, sodium, potassium, and/or BUN range for a user wearing device. For example, the appcan provide a graphical plot of one or more values throughout a time period, for example a day. This graphical plot and/or view may also mark hyper- and hypo-glycemic boundaries. In some embodiments, the appsmay provide a number to serve as a surrogate for good glycemic control and set targets for the user to help achieve specific glycemic control ranges. These targets can be set in an incremental way such that the user is able to gradually change behavior to achieve realistic and lasting lifestyle changes. In some embodiments, the appsmay be synchronized with a third party health app to provide users a holistic view of user-based glucose, metabolites, diet, and/or activity, just to name a few examples.

204 102 104 228 209 210 212 208 204 228 204 204 226 In general, the appsmay be used by a user of the wearable deviceto interface with output from ML models, sensors, generative system, prediction system, image analyzer, and mapping logic. In some embodiments, at least one of the appsmay provide a trigger for a sensor (e.g., sensor) to capture data, send the captured data to an online storage medium (e.g., a cloud server) for ML processing and/or prediction system processing, and push the predicted results back to the app. The appmay then display the predicted results on the optional display, and/or may display results on another computing device display.

204 108 204 102 102 108 In some embodiments, the appsmay include an AI application for coaching or assisting a user with lifestyle choices. For example, the lifestyle AI coachmay execute as an appon deviceand may utilize data from the wearable device, user inputs, and external sources to provide personalized guidance. The AI coachmay analyze patterns in glucose levels of the user, correlating them with diet, exercise, sleep, and stress data to identify factors influencing glucose management. Based on this analysis, the coach may generate tailored recommendations, such as suggesting specific meal compositions to minimize post-prandial glucose spikes or recommending optimal times for physical activity to improve glucose control.

108 108 108 108 In some embodiments, the AI coachmay also adapt to a selected or predefined communication style and frequency based on user preferences and engagement levels. For example, the AI coachmay use motivational interviewing techniques to help users overcome barriers to behavior change, asking open-ended questions and providing reflective feedback. The AI coachmay employ different strategies for different users. For example, the AI coachmay use gamification elements for those motivated by competition, or may provide detailed scientific explanations for users that prefer to understand the underlying mechanisms of glucose regulation.

108 108 108 108 In some embodiments, the AI coachmay anticipate potential challenges and proactively provide strategies to maintain healthy behaviors for a user. For example, if a calendar associated with the user indicates an upcoming business trip, the AI coachmay offer tips for managing glucose levels while traveling, including restaurant recommendations and strategies for staying active. The AI coachmay also integrate with other apps and services, such as suggesting recipes that align with user health goals and may, for example, automatically add ingredients to a shopping list. By providing continuous, personalized support, the AI coachaims to make sustainable lifestyle changes more achievable for users managing glucose levels.

200 200 In operation of system, analysis performed may be representative of particular particles, elements, or glucose flowing through the veins. For example, the systems described herein may determine a concentration of glucose using an optical sensor, such as an NIR optical sensor positioned substantially adjacent to a body site of a user. In some embodiments, images captured by the image sensors described herein may be analyzed by systemto determine correlations between changes in blood glucose levels by assessing changes in the appearance of the veins in the captured images.

200 102 200 200 In some embodiments, the system(e.g., wearable device) may determine a concentration of urea by analyzing image data captured by the NIR optical sensor (or another onboard sensor). For example, the systemmay detect changes in the optical properties of the venous structures that correlate with urea concentration. In this instance, specific NIR spectrum wavelength(s) may be utilized to determine the absorption characteristics of urea exist and determine intensity of the light. This information can be used in conjunction with determined glucose data to provide a comprehensive assessment of the metabolic state of the user. The systemmay alert the user if urea levels indicate potential kidney function issues or dehydration, prompting (e.g., nudging) the user to increase fluid intake or seek medical advice.

200 228 In a non-limiting example, the systemmay detect changes in a darkness of the veins, which may be attributed to the NIR light interactions. The higher the glucose concentration, the lower the scattering coefficient of NIR photons and the darker the veins, corresponding to fewer photos returning back to the sensor. The changes in the size and volume of the veins may be attributed to the osmolality changes. The higher the glucose concentration, the higher the osmolality and larger the size or volume of the veins, corresponding to water flow into the veins.

200 102 108 In some embodiments, the systems may determine a concentration of sodium by analyzing the image data captured by the NIR optical sensor (or another onboard sensor). For example, the system(e.g., wearable device) may detect subtle changes in tissue fluid balance and blood vessel characteristics that correspond to sodium levels. Users can use this information to manage hypertension and/or heart conditions alongside diabetes or prediabetes conditions. The AI coachmay use the data to provide tailored dietary recommendations, such as suggesting low-sodium alternatives or advising on optimal hydration strategies based on the detected sodium levels and activity patterns of the user.

200 102 108 In some embodiments, the systems may determine a concentration of chloride by analyzing image data captured by the NIR optical sensor (or another onboard sensor). For example, the system(e.g., wearable device) may identify osmolality changes associated with chloride ions in the bloodstream. In some embodiments, particular NIR optical sensors may be utilized for particular determinations. For example, for chloride ion determination a suitable porphyrin based sensor such as 5, 10, 15, 20-Tetrakis (4-hydroxyphenyl) porphyrin (THPP) may be used. Chloride levels can also provide insights into an electrolyte balance and hydration status of the user. The AI coachmay use this information to fine-tune recommendations for fluid intake. This can ensure that users engaged in intense physical activity or those living in hot climates are provided a recommendation based on context of the body engagement and location.

200 102 108 In some implementations, the systems described herein may determine a concentration of potassium using the NIR optical sensor (or another onboard sensor). For example, the system(e.g., wearable device) may detect changes in the electrical properties of tissues that correlate with potassium levels. For example, utilizing a specific NIR spectrum wavelength (e.g., spectral range between about 900 nm to about 1700 nm) sensitive to the presence and concentration of potassium. This measurement may provide a potassium monitoring mechanism for users taking certain medications that affect potassium balance. The AI coachmay provide alerts if potassium levels approach concerning thresholds and may offer dietary suggestions to maintain optimal potassium levels, such as recommending potassium-rich foods when levels are low or advising caution with high-potassium foods when levels are elevated.

200 102 108 In some embodiments, the systems described herein may determine a concentration of bicarbonate based on analyzing image data captured by the NIR optical sensor (or another onboard sensor). For example, the system(e.g., wearable device) may detect subtle changes in blood pH levels reflected in the optical properties of venous structures. For example, utilizing a specific NIR spectrum wavelength (e.g., spectral range between about 700 nm to about 2500 nm) sensitive to the pH levels. Bicarbonate levels can provide insights into an acid-base balance in the body, which can help to assess overall metabolic health. The AI coachmay use this information to provide guidance on dietary choices that support optimal pH balance, such as suggesting alkaline-forming foods if bicarbonate levels are low, or advising on strategies to mitigate metabolic acidosis in conjunction with glucose management.

2 FIG. 212 228 102 212 102 102 228 Referring again to, the image analyzermay receive images (e.g., image data) from one or more image sensors(e.g., an NIR sensor). The received images may include data representing one or more digital images capturing a body site of a user wearing device. The image analyzermay assess the images to measure and/or detect both physical and physiological changes under human skin, such as within veins and in particular, venous blood. The physical and/or physiological changes in a vein responsive to IR light, for example, may be used by deviceto ascertain changes in glucose for the user wearing device. For example, utilizing a specific NIR spectrum wavelength (e.g., spectral range between about 700 nm to about 2500 nm, about 900 nm to about 2000 nm, about 1000 nm to about 1800 nm, etc.) sensitive to the glucose molecules. For example, the changes in the darkness of one or more veins at the body site may be attributed to IR light interactions between the sensorand the body site. Changes in the size and/or volume of the one or more veins at the body site may be attributed to osmolality changes.

In general, osmolality represents a colligative property of a solution (including body fluid) that indicates the concentration of all the particles dissolved in the solution. Assessing osmolality using IR light, as described herein, can allow for determination of glucose changes in venous blood within a vein at a body site. In human blood serum or blood plasma, osmolality depends on the concentration of five major osmotic solutes. Three of the five osmotic solutes are of ionic nature (i.e., Na+ (sodium ion), Cl— (chloride), HCO3- (bicarbonate)). The final two osmotic solutes are non-ionic (i.e., glucose and urea). Water normally flows from a compartment of low osmolality to a compartment of high osmolality and this occurs if a membrane between the two compartments is permeable to water. When water moves between plasma and intracellular compartments, the movement direction depends on the osmolality of both compartments. For example, if a cell is in a relatively hyperosmolar solution, fluid will move out of the cell towards the highly concentrated compartment to reach homeostasis. As a result, the cell will shrink. There are many ways to estimate serum osmolality utilizing the basic metabolic panel test to measure sodium, glucose, potassium, and/or blood urea nitrogen.

−1 −1 4 12 FIGS.A-D In addition, water absorption in the shorter NIR wavelengths is typically lower (e.g., about 0.004 mmto about 0.04 mm). Tissue penetration in the shorter NIR wavelengths is typically higher (e.g., about 2 mm to about 3 mm). Both together make this wavelength range suitable for sensing in humans. Clinical measurements corresponding to embodiments shown inindicate a good correlation between reduced scattering coefficients of intracellular mediums and extracellular mediums in the NIR range and the concentration of glucose as determined from a conventional blood test of glucose.

102 102 The majority of glucose in the human body is found within interstitial fluid (i.e., between cells), rather than being contained in blood. Accordingly, the embodiments described herein may discern a location of one or more veins, determine where the one or more veins are located in relation to surrounding tissue, and perform analysis on changes over time in such veins and surrounding tissue. In addition, water also has a strong absorption spectrum in the NIR wavelengths. However, interference from water may be avoided by the embodiments described herein due to the ability of the wearable deviceto discern veins from other tissues of the body. Further, under normal bodily function circumstances, the concentration of water in venal blood remains steady and thus is unlikely to interfere with assessments performed by the deviceon captured images of the one or more veins.

212 104 104 212 212 104 104 Since glucose levels fluctuate immediately with food intake but sodium levels do not typically fluctuate immediately with food intake, changes in osmolality may predominantly reflect changes in blood glucose. Thus, the image analyzercan obtain and analyze images and/or image data and provide such image data and analysis to one or more ML models. In some embodiments, one or more ML modelscan preprocess (e.g., resizing, rescaling, sampling, cropping, rotating, flipping, etc.) the images before the image analyzeranalyzes the images. In some embodiments, the image analyzeris a module within an ML modeland may function as an image analyzer algorithm for the ML model.

209 104 250 251 251 209 210 104 102 1500 1550 209 209 209 The generative systemmay function with any or all ML modelsto assess user behavior and user indications, for example, as a basis in which to generate recommendationsfor improving such user behavior with respect to disease states. The recommendationsmay be based on observations and/or predictions generated by system, system, or other modelassociated with device, band, band, or the like. In some implementations, the recommendations may be based on observations assessed via other devices or input by the user at a computing device such as a smart phone. The generative systemmay represent an AI model that is trained to generate content based on prompts to the model. The systemmay receive a user input requesting a text output from the model, and the model may generate and output text based on the user input. Systemmay be referred to as a question and answer model, where the model is provided a natural language question and the model generates a natural language answer to the question.

209 209 200 209 In some embodiments, the generative systemmay allow for multi-shot prompting. A standardized format or output template may be used depending on the content being generated. For example, if systemis generating a proposed grocery list, the systemmay automatically provide particular stored templates for presenting grocery list suggestions. By contrast, a user may request different formats and/or templates when requesting a grocery list. The templates may be predefined or user defined. In some embodiments, the template may mimic output provided in a previous request when the requests are similar in nature. For example, rather than following a predefined template, the generative systemmay be aware of a prior grocery list request and may provide new output for a different grocery store, but may choose to provide the new output according to the previous output format, alignment, or other similar textual or graphical feature. Providing output in a predictable format provides an advantage to users of expected form and function with a recognizable user interface. Such an advantages enables a user to understand and engage with the output at an accelerated pace than if the format were to differ between queries. Moreover, using a similar or previously known format can improve the ability of the user to determine the reliability and/or accuracy of the output based on the recognition of the way similar data had been previously presented.

210 104 210 102 250 251 210 210 In general, the prediction systemmay function with ML modelsto preprocess, optionally segment, analyze, and predict blood glucose levels within one or more venous blood vessels. For example, the prediction systemmay generate one or more outputs such as glucose levels, glucose measurements, glucose predictions, and/or related predictions. In some embodiments, the prediction system may generate the one or more outputs and wearable devicemay use such outputs to generate indicationsand/or recommendations. For example, prediction systemmay be used to predict sodium, potassium, glucose, and/or BUN since serum osmolality is based on these metabolites. Prediction systemmay use one or more of the following algorithms to predict sodium, potassium, glucose, and/or BUN.

104 104 104 104 104 In some embodiments, ML modelsmay include a plurality of layers. For example, one or more shallow or lower layers may learn simple features, and one or more deeper layers may learn more complex features. In one exemplary, non-limiting embodiment, a first layer may learn simple features like edges (vertical, horizontal, diagonal) or shades of color. A second layer may learn more complex features such as shapes and textures. A third layer may learn further complex patterns like honeycomb shapes, multi-layered or dimensional shapes, etc. A fourth layer may be able to detect whole objects. A fifth layer may learn features within whole objects or associated with whole objects. Employing exemplary ML modelsto the embodiments described herein may include using one or more lower layers to detect edges of veins, sizes of veins, shapes of veins, vein or skin color contrast, etc. Further, one or more deeper layers may be used to learn a wrist, background, fingers, whole arm, etc. Since the ML modelwould be able to learn the shapes of veins, color contrast, and the size of the veins, the ML modelmay learn to detect changes in the size, shape, and/or color of the veins. In other words, the ML modelmay detect osmolality changes associated with one or more veins. In some embodiments, one or more connected layers may classify the various changes in size, shape, and/or color into probabilities of each glucose range, as described elsewhere herein.

250 Indicationsmay be provided and/or presented as text, audio data, and/or visual data including, but not limited to: observations, observation periods, glucose statistics, severity levels, raw measurements, predictive events, and predictive levels of glucose, etc. In some embodiments, the data may further include treatment options, diagnoses, activity levels, patterns in glucose over time, and/or other fitness or health measurements. For example, data may include reports with detailed analyses of glucose level predictions as well as various treatment options for a specific user. In some embodiments, alerts, messages, and/or reminders about particular glucose testing (e.g., image capture) and glucose assessments (e.g., data and/or reports that are generated on a schedule) may notify the user with a visual or audial cue.

251 102 1500 1550 Recommendationsmay be provided and/or presented as text, audio data, and/or visual data including, but not limited to: observations, observation periods, behavior changes, behavior patterns, purchase recommendations, glucose statistics, severity levels, raw measurements, predictive events, and predictive levels of glucose, etc. In some embodiments, the data may further include treatment options, diagnoses, activity levels, patterns in glucose over time, and/or other fitness or health measurements. For example, data may include reports with detailed analyses of behavior (e.g., eating and/or exercise events or patterns), glucose level predictions as well as various treatment or behavior recommendation options for a specific user. In some embodiments, alerts, messages, and/or reminders about particular user behavior may notify the user with a visual or audial cue via device, band, band, or the like.

210 104 104 212 102 210 102 228 102 The prediction systemmay be part of ML modelsand may utilize ML modelsand/or image analyzerto generate predictions of glucose levels over time based on captured images and/or reflections from a body site of a user wearing device, for example. In some embodiments, the prediction systemmay determine or predict blood glucose measurements/estimates of a user wearing deviceby using dynamic image attributes of a body site of the user from one or more sensorson device.

210 104 102 210 In some embodiments, the prediction systemrepresents a diabetes classification system that may predict a likelihood of a user having and/or developing prediabetes, T2D, or another blood sugar disorder based on classifying glucose levels identified by ML modelsby assessing a historical selection of images of venous blood vessels of a subject (e.g., user) wearing device, for example. In some embodiments, the prediction systemmay further provide predictions for the user with respect to health decline or health improvement based on glucose levels that are determined over time for a particular user.

250 204 102 102 102 Predictions may be provided and/or presented as a report including data such as observations, observation periods, glucose statistics, severity levels, raw measurements, etc. In some embodiments, report data may further include treatment options, diagnoses, activity levels, patterns in glucose over time, and/or other fitness or health measurements. For example, reports may include detailed analysis of predictions as well as various treatment options for a specific user. In some embodiments, alerts, messages, and/or reminders about particular glucose testing (e.g., image capture) and glucose assessments (e.g., reports that are generated on a schedule) may notify the user with a visual or audial cue. For example, a visual cue may refer to a message indicator (e.g., indication(s)) in an appor an email in an email app. An audio cue may be provided as an audio output from a speaker of device, an audio output from a digital assistant in communication with device, and/or an audio output from other computing device in communication with device.

104 102 104 208 104 228 104 208 In some embodiments, the ML modelsmay map the detected changes in the blood to particular glucose levels. The devicemay obtain additional glucose signals over time by monitoring the body site. Data may be generated from the monitoring and the ML modelsmay map changes over time. In some embodiments, the mapping logicmay be used in combination with ML modelsto map the detected changes in blood to particular glucose levels. For example, one or more images may be obtained from sensor. The one or more images may visually show a correlation between a change in a glucose level associated with one or more veins (captured in the one or more images) and a change in an appearance (e.g., a darkness, a size, and a volume) of the one or more veins. The visual changes in the vein can be mapped by ML modelsusing mapping logicto map the visual features/changes to known glucose values.

208 208 208 The mapping logicmay represent a set of rules for mapping visual features/changes of a particular image of venous blood vessels to predefined glucose values. For example, the mapping logicmay include one or more data structures that correlate blood glucose levels with physical and/or physiological features of a venous blood vessel. The one or more data structures may include one or more lookup tables, algorithms, or techniques. The mapping logicmay be applied to function on pixels, voxels, circles, ovals, rectangles, squares, and/or other polygon or predefined shape.

208 In an example, non-limiting embodiment, mapping logicmay include using an edge detector algorithm to detect the edges of one or more blood vessel portions in one or more images; segmenting image portions of the one or more blood vessel portions based on the detected edges; aligning the segmented image portions over time or from two or more time periods; determining a blood volume change between a first image, at a first time, showing the one or more blood vessel portions and a second image, at a second time, showing the one or more blood vessel portions; detecting, using a contrast detector, a change in the darkness or intensities of the pixels between a first image, at a first time, showing the one or more blood vessel portions and a second image, at a second time, showing the one or more blood vessel portions; and mapping the change in darkness or intensities to a corresponding change in a glucose level done based on the determined changed in blood volume and the determined change in darkness or intensities of the pixels. In some embodiments, the first time and the second time are after a predefined event, event-based, on-demand, otherwise. The events may include, but are not limited to, one or more of: a meal, a fast, a physical stressor, exercise, an emotional stressor, a sleep event (e.g., excess sleep, sleep deprivation, exhaustion, etc.), a time of day, and a change in estrogen.

228 102 104 102 These changes in an appearance (e.g., darkness, size, and volume) of the one or more veins, as captured by the sensor, can be ascertained by the deviceand ML modelsbased on a combined effect of the reduced scattering coefficient of NIR light in the presence of glucose molecules and the change in plasma osmolality as described in detail herein. Determining changes in the appearance of one or more veins of a user wearing device, for example, can be used as a way to measure venous blood glucose values for the user in a noninvasive way by using image capture and machine learning image assessments.

104 208 104 104 104 104 104 104 In some embodiments, the ML modelsmay utilize mapping logicto map detected changes in the appearance of veins before and after events. The events may include, but are not limited to one or more of: a meal, a fast, a physical stressor, exercise, an emotional stressor, a sleep event (e.g., excess sleep, sleep deprivation, exhaustion, etc.), a time of day, and a change in estrogen. In some embodiments, using ML modelsto map detected changes in the appearance of veins before and after events allows the learned patterns and predictions (generated by ML models) about glucose determinations and/or related predictions to be incorporated into a semi-supervised ML learning stage of ML models. For example, a semi-supervised stage of an ML model may include learning from both unlabeled and labeled samples, typically sampled from the same or similar distributions. In some embodiments, using ML modelsto map detected changes in the appearance of veins before and after events allows the learned patterns and predictions (generated by ML models) about glucose determinations and/or related predictions to be incorporated into a self-supervised ML learning stage of ML models.

104 In some embodiments, one or more of ML modelsmay employ self-supervised learning. Self-supervised learning uses surrogate (pretext) tasks that can be formulated using unsupervised data. A pretext task is designed in a way that solving the pretext task includes learning of a useful image representation. In some of the embodiments described herein, the pretext task may be learning representations by maximizing agreement between different views of the veins of each user obtained at the same instant of time. For example, the process may include supervised pretraining using large scale non-medical labeled data (e.g., ImageNet). The process may further include multi-instance contrastive self-supervised pretraining using domain specific unlabeled data, resulting in learning a representation or pretext task. For example, an unlabeled dataset may be collected from any of the devices described herein without a clinical study because the data is unlabeled (i.e., does not include corresponding venous blood glucose values). The process may further include supervised fine-tuning using labeled dataset collected from, for example, a clinical study. This may be the classification step that learns to output glucose range, glucose number, a change in glucose, a glucose color, or the like.

206 102 206 Patient datamay include, but is not limited to, patient identifying information (e.g., a name, birthdate, etc.), patient-entered data, and digital data indicative of historical and/or current conditions of a patient, diagnostic tests, and test results associated with the patient, medications prescribed or administered to the patient, therapies (e.g., surgeries, outpatient procedures, or the like) administered to the patient, prior and real time glucose predictions and measurements captured by wearable device, data stored or extracted from an electronic medical record (EMR), demographics of a patient, vital signs of a patient, notes from one or more medical appointments attended by the patient, or any combination thereof. In general, patient datamay be used by the systems and processes described herein when permission for using such data has been provided by the patient.

218 102 218 204 104 212 208 218 One or more processorsmay be included in device. The processorsmay be used for executing program instructions associated with apps, ML models, image analyzer, and/or mapping logic. In some embodiments, the processorsmay include graphical processing units (GPUs) for rendering and generating graphics, images, video, audio, and/or multimedia files.

220 222 102 220 222 Other circuitry such as an analog-to-digital converter (ADC)and a digital-to-analog converter (DAC)can also be located within wearable device, such as co-integrated as a portion of a commonly-shared integrated circuit or integrated circuit package. The ADCmay convert an analog signal representation of a captured image into a digital representation of the image. The DACmay convert a digital signal representation of a captured image into an analog signal for further processing.

210 212 208 104 102 210 212 208 104 106 102 212 104 212 104 210 104 210 104 Although the prediction system, image analyzer, mapping logic, and the ML modelsare depicted as part of the wearable device, any or all of such functionality carried out by the prediction system, image analyzer, mapping logic, and/or the ML modelsmay alternatively, or additionally, be carried out by an external computing device, such as the computing devicein wired or wireless communication with wearable device. In addition, while the image analyzeris depicted external to (e.g., not integrated into) ML models, in some embodiments, the image analyzermay be integrated into the ML models. In addition, while the prediction systemis depicted internal to (e.g., integrated into) ML models, in some embodiments, the prediction systemmay be external to (e.g., not integrated into) the ML models.

102 240 242 244 240 102 228 240 102 The wearable devicealso includes or is communicatively coupled to power, input devices, and output devices. The powermay be a power source for the wearable deviceand/or sensors. The powermay include or connect to a battery or a port for connecting the wearable deviceto a power supply, a wall power adapter, a battery pack, or the like.

242 202 218 228 204 212 210 104 208 206 242 102 242 102 204 224 102 242 The input devicesmay interact with memory, one or more processors, sensors, apps, image analyzer, prediction system, ML models, mapping logic, and/or patient data. The input devicesmay include touchscreens, switches, buttons, toggles, and/or other hardware components located on wearable device. In some embodiments, the input devicemay be external to or not integrated into the device, such that one or more controllers, mobile device, apps (e.g., apps), etc., may communicate (e.g., one or more optional antennas, coils, etc.) with deviceusing a wireless communications protocol. In some embodiments, the input devicemay include, for example, a touch input device that can receive tactile user input, a microphone that can receive audible input, and the like.

244 202 218 228 204 212 210 104 208 206 244 226 The output devicesmay interact with memory, one or more processors, sensors, apps, image analyzer, prediction system, ML models, mapping logic, and/or patient data. The output devicesmay include, for example, a display (e.g., optional display) for visual output, a speaker for audio output, and/or haptic feedback.

102 102 228 232 228 228 102 218 In some embodiments, the wearable deviceis a noninvasive blood glucose monitoring system that is configured to be removably coupled to a body site of a user. The wearable devicemay include at least one optical sensor (e.g., image sensor) communicatively coupled to at least one light source (e.g., red-light source, near infrared light source, optional light sourceor a light source internal to sensor). The optical sensormay sense reflected light from the body site resulting in a captured image. The wearable devicemay also include and be communicatively coupled to at least one processor (e.g., processor) which may be programmed with the techniques and algorithms described herein to programmatically analyze the captured image to determine a blood glucose estimate for the user.

200 200 209 209 210 210 b b In some embodiments, systemmay function to nudge users toward improved health behaviors, particularly in individuals exhibiting prediabetes or risk factors for developing type 2 diabetes. The systemmay include one or more generative AI modules, such as large language models (LLMs, generative systems,), one or more predictive models (e.g., predictive systems,) that may function to provide personalized dietary, physical activity, and behavioral guidance based on user-specific health status and preferences.

220 251 102 The nudges described herein may represent indicationsthat may represent recommendationsor other prompts to enable a user of wearable deviceto improve health metrics or lifestyle metrics. Example nudges may include dietary behavior modifications, nutrient modifications, behavioral modifications, exercise modifications, stress modifications, and/or sleep modifications.

200 200 In a non-limiting example, the systemmay identify a user with prediabetes or diabetes risk factors and generate a personalized behavioral intervention plan. The systemmay recommend a targeted weight reduction of approximately 7% of total body weight through improved diet quality and increased physical activity, for example.

200 251 200 104 104 200 The systemmay generate recommendationsfor dietary behavior modifications. For example, the systemmay identify high added sugar consumption as a modifiable risk factor. One or more ML modelsmay generate tailored recommendations that prompt the user to eliminate sugar-sweetened beverages (e.g., soda, candy, fruit drinks, etc.), informing the user with additional education. For example, the ML modelsmay further indicate to the user that such items constitute empty calories with minimal nutritional value. The systemmay further query the user to identify frequently consumed sugary items and generate behavioral nudges that frame these foods as occasional rewards (e.g., consumed once weekly instead of daily).

200 200 The systemmay also assess an intake of refined carbohydrates (e.g., white bread, white rice, pastries, etc.) and recommend substitutions with whole grain options, such as whole wheat, barley, millet, and oats. In some embodiments, the systemmay provide an explanation regarding the limited nutritional value of enriched flours as compared to whole grains, including the loss of dietary fiber.

102 104 200 200 104 109 104 In cases where devicedetects glycemic spikes despite whole grain consumption, the ML modelmay suggest pairing carbohydrates with proteins (e.g., egg) or healthy fats (e.g., almond butter) to moderate glycemic response. If a user is shopping or indicating a request for a shopping list, the systemmay be also analyze food labels for added sugars and recommend specific brands that meet nutritional criteria, alleviating the burden of label analysis. For example, the systemmay receive a user prompt such as, ‘Which whole grain breads do not contain added sugar?’ and the ML modelmay return product-specific recommendations with annotated nutritional information. A secondary software agent (e.g., agent) may be called by the ML modelto extract label data, calculate added sugar or calorie totals, and/or compare these values against thresholds defined by health guidelines (e.g., 24 grams/day for adult women).

200 251 200 200 251 200 200 The systemmay generate recommendationsfor fiber intake enhancement. For example, the systemmay identify insufficient dietary fiber as a common deficiency and recommend increased intake through fruits, vegetables, and cereal-based sources (e.g., wheat, oats, millet, etc.). In some embodiments, the systemmay provide educational content along with the recommendations. For example, the systemmay provide indications in a user interface that explain the roles of soluble and insoluble fiber in cholesterol and blood glucose management, respectively, and may further suggest whole foods as the preferred delivery vehicle. The systemmay also provide optional guidance on fiber supplementation when applicable.

200 251 200 200 104 The systemmay generate recommendationsfor alcohol intake monitoring. For example, in some implementations, the systemmay prompt users to assess personal alcohol consumption frequency and context (e.g., social functions, weekends). The systemmay suggest behavioral replacements for alcoholic beverages, including sparkling water, kombucha, or non-alcoholic alternatives. The ML modelmay also dispel outdated health beliefs regarding alcohol (e.g., the heart-health benefits of red wine), citing recent evidence attributing such benefits to the overall dietary pattern rather than alcohol consumption itself.

200 251 200 The systemmay generate recommendationsfor temporal eating patterns. For example, to further support glycemic control, the systemmay guide users to establish a cutoff time for evening eating (e.g., 8 or 9 PM), thereby minimizing late-night glucose spikes. Additional example behavioral guidance may include instructions to consume proteins and vegetables prior to carbohydrates during meals, which can modulate postprandial glucose levels.

200 251 200 104 The systemmay generate recommendationsfor physical activity integration. For example, the systemmay prompt users to identify preferred forms of physical activity. Both aerobic and strength-based exercises are recommended, as each contributes to increased insulin sensitivity. The ML modelmay offer suggestions ranging from moderate-intensity activities (e.g., walking, cycling) to vigorous exercise (e.g., running, lap swimming), and remind users that consistency outweighs selecting a specific modality.

200 251 200 200 104 The systemmay generate recommendationsfor stress detection and management. For example, the systemmay monitor indicators of user stress and may invoke a set of actions to reduce stress-related glucose elevation. If stress is detected, the systemmay activate a game-based nudging mechanism and/or a humor-driven nudging mechanism. A game-based nudging mechanism may launch a gamified behavioral experience, such as a digital version of a game in which beneficial food choices (e.g., vegetables) allow users to ascend toward winning the game while poor choices (e.g., candy) result in setbacks (e.g., descending a feature of the game). A humor-driven nudging mechanism may include delivering context-aware humorous content, such as stand-up comedy clips or ML model-generated jokes, to elicit laughter and reduce cortisol levels. In some embodiments, selection of the nudging mechanisms may be user specific. For example, a younger user may enjoy digital games and be more engaged with a gamified nudging.

200 The AI/ML models described herein can be prompted to suggest all of the above and in that order based on training, user behavior detection, user input, and the like. Moreover, the systemmay provide real-time user monitoring of diet and activities over time and provide such information in a stepped fashion to ensure compliance to steps for improving behavior, to build knowledge slowly, and to provide research-based medical knowledge on an as detected (or as requested) basis.

200 108 108 In a non-limiting example, the systemmay adapt recommendation generation behavior and recommendations based on various contextual factors. For example, if a user is attending a conference and unable to adhere to typical recommendations, the lifestyle AI coachmay inquire about the conference details. The AI coachmay utilize a web-searching function to obtain information about the conference end date and may adjust a nudging mode or strategy, accordingly, thereby resuming normal recommendations after the conference concludes.

108 102 200 200 1 In some embodiments, a nudging strategy may represent any number of nudging modes. For example, the AI coach(or user of wearable device) may select a light mode, a regular mode, or a heavy mode to receive nudges generated based on the selected mode. By way of example, the systemmay account for travel scenarios, which often disrupt user health goals. During travel periods, the nudges may transition to the light mode, maintaining minimal guidance to ensure users still consume adequate fiber, vegetables, and water without feeling overly restricted. In general, the systemmay implement different modes of nudging intensity, including light mode, regular mode, and heavy mode. Light mode may be activated during holidays, business, or personal travel, the postpartum period, short-term illnesses or injuries (personal or of loved ones), and user-designated cheat/light days. Conversely, heavy mode may be employed after discontinuation of certain medications like GLP-agonists to prevent rapid weight regain, or when users set aggressive goals, such as in preparation for an upcoming doctor visit.

108 108 200 1 108 The lifestyle AI coachmay provide various types of recommendations tailored to a particular user context. Example recommendations may include tips on meal timing, focusing on earlier eating and monitoring blood glucose during time-restricted eating trials. The AI coachmay also offer guidance on food choices and meal sequencing, assessing glucose control in the context of specific diet plans such as ketogenic diets. Example exercise recommendations may cover both type and duration of physical activities. The systemmay suggest recipe modifications, such as incorporating more legumes or vegetables into pasta dishes, and may provide strategies for managing food cravings. In examples in which users are discontinuing a GLP-agonist medication, the AI coachmay offer specific advice for managing withdrawal symptoms and maintaining blood sugar control, including suggestions for nutrient-dense foods and monitoring calorie and/or fiber intake.

108 200 108 A tone and approach of the lifestyle AI coachmay be dynamically adjusted based on user feedback. For example, the systemmay implement a quick poll using emojis to gauge a current emotional state of the user, thereby allowing the AI coachto generate a more empathetic or assertive approach accordingly. This emotional input may also influence the selection of nudging modes.

108 200 108 108 In some embodiments, the lifestyle AI coachmay provide location-specific suggestions, such as identifying nearby grocery stores or recommending healthy menu options from local restaurants. The systemand/or lifestyle AI coachmay account for seasonal or cultural events that may impact eating patterns. For example, if consistent nighttime glucose spikes are detected for a user during Ramadan, the AI coachmay adjust recommendations and shift to a light mode of nudging until the end of the fasting period. Similar adaptations may be made for other holidays that follow non-solar calendars, such as Diwali or Hanukkah.

200 108 The systemmay leverage milestones such as birthdays to encourage behavior change, prompting users to resume glucose monitoring if frequency has decreased and/or suggesting the setting of new lifestyle goals. Subsequently, the AI coachmay implement quarterly check-ins to remind users of these goals and assess progress.

200 108 200 200 108 In some embodiments, the systemmay implement behavior-based adjustments. For example, if the lifestyle AI coachdetects systematic elevations in glucose levels over the course of a month, the systemmay provide nudges reminding the user of healthful meal components and encourage the reintroduction of routine exercise. This approach helps address the common issue of declining motivation between healthcare provider appointments, using device-generated data to re-motivate users even before clinical testing occurs. The systemmay also account for the tendency of users to relax effort after a period of successful outcomes. To address this, the nudge mode may be switched to a light mode after a prolonged stretch of success, such as 3-6 months, and then may revert to the previous mode if the AI coachdetects a decline in adherence to health goals.

200 108 The frequency and intensity of nudges may be adjusted based on a glucose time in range (or the proportion of glucose values within normal limits) for the user. If the time in range is consistently high, the systemmay reduce monitoring frequency to weekly checks. However, if values begin to fall out of a predefined range (described elsewhere herein) more often, the AI coachmay adapt by deploying more frequent reminders on lifestyle changes and daily monitoring until glucose levels improve. In some embodiments, users may select reminders and monitoring schedules, for example, some users may become used to receiving reminders and monitoring and thus may appreciate the consistency.

108 108 In some embodiments, the lifestyle AI coachmay also provide real-time, meal-specific guidance. For example, if a trend towards increasing blood glucose levels is detected after typical mealtimes, the AI coachmay prompt the user to consume protein-containing foods first at the next meal and suggest a particular meal sequence based on the user entering input including upcoming, planned food items.

108 200 108 In some embodiments, the lifestyle AI coachmay provide exercise recommendations that are fine-tuned based on glucose data collected after different types of physical activity. The systemand/or the lifestyle AI coachmay curate personalized exercise programs by analyzing glucose responses to various forms of aerobic and resistance training.

102 108 108 In some embodiments, the wearable devicemay be made aware of circumstances such as pregnancy. In response, the lifestyle AI coachmay adapt guidance and/or recommendations to address specific health concerns like appropriate weight gain. The AI coachmay also provide tailored advice for the postpartum period, when users often shift focus away from personal health and towards infant care.

108 108 200 200 17 18 FIGS.A-B To reduce barriers to healthful eating, the lifestyle AI coachmay inquire about which grocery store(s) the user wishes to shop. Using web-searching or maps-searching functions, the AI coachmay suggest alternative stores in the neighborhood if a usual choice is suboptimal from a nutritional standpoint. The systemmay also provide specific food recommendations within a given store, suggest quick and healthy recipes, and provide detailed shopping lists including ingredient quantities and potential store locations. By making healthful eating more convenient and accessible, the systemaims to facilitate easier adoption of beneficial dietary habits. Additional examples are described in.

200 108 108 200 109 109 108 204 In a non-limiting example, the systemmay address financial constraints that may impede healthy lifestyle choices. For example, the lifestyle AI coachmay inquire about user budget constraints. Upon receiving this information, the AI coachmay employ a web-searching function to locate nearby grocery stores and browse through healthy food options. The systemmay then activate a coding agent (e.g., agent model) to generate code or call a predefined function to identify favorable pricing on services and/or food items. The agent modelmay generate an algorithm that aggregates the prices of healthy items and ensures that the total cost remains within the specified budget limit. Based on this analysis, the lifestyle AI coachmay present the user with a list of budget-friendly, healthy food options. To further illustrate the financial benefits of home cooking, the appmay provide a chat interface to perform a cost comparison between takeout meals and home-cooked alternatives, displaying the potential financial incentives for preparing meals at home.

200 200 200 200 200 200 In some embodiments, the systemmay account for loss aversion or bias with respect to understanding that some users may wish to work harder to avoid losses rather than work the same amount to achieve equivalent gains. For example, the systemmay enable a user to avoid future losses. In a non-limiting example, the systemmay determine to provide a takeout meal versus home cooked meal price comparison and financial incentive to encourage selection of the home cooked meal. For example, the systemmay specifically calculate an amount of money to be saved by allowing the systemto suggest and perform grocery shopping for a home cooked meal. In some embodiments, the systemmay further provide a recipe to encourage the user to choose a home cooked meal.

200 200 200 200 200 In some embodiments, the systemmay account for loss aversion bias of a user. For example, the systemmay calculate and depict progress for accomplishing behavioral goals to encourage continued goal increasing behavior. For example, the systemmay set goals for the user such as perform moderate physical activity 3 times per week and/or limit added sugars for 5 days in a row. The systemmay monitor user behavior to assess for the set goals and may project changes in endurance or sugar levels or other health value when detecting a goal failure or backward slide from a set goal. Such goals may be overlaid with other rules, goals, and interruptions. For example, a holiday or planned exercise hiatus can be accounted for with the set goal because the systemmay extend the days or per week goal to exclude the planned downtime.

200 108 The systemmay also address time constraints, which are often cited by users as a barrier to adopting healthier lifestyles. For example, the lifestyle AI coachmay assist users in efficiently restocking food items at home, potentially saving time typically spent on grocery shopping and meal planning. This feature may be beneficial for users following specific dietary guidelines, such as those recommended by diabetes prevention programs, which can be time-intensive to implement.

108 200 108 To combat a lack of immediate symptoms often associated with elevated glucose levels, the lifestyle AI coachmay be programmed to educate users about the long-term implications and potential consequences of developing diabetes. The systemmay employ strategies to counteract optimism bias, which is the tendency to underestimate the likelihood of negative events and overestimate the likelihood of positive ones. The lifestyle AI coachmay also provide immediate feedback on glucose levels, interpreting data and presenting typical rhythms of blood glucose changes without having the user read and interpret complex charts. This real-time feedback may serve as motivation for making healthier choices in subsequent meals or during the next grocery shopping trip.

200 108 In some embodiments, the systemmay incorporate gamification techniques to leverage present bias, which is the tendency to assign greater value to immediate rewards compared to future benefits. For example, the AI coachmay implement a reward system for achieving exercise goals or improving eating habits, providing immediate positive reinforcement for healthy behaviors. In some embodiments, the rewards may include any suitable goal for a user, such as foods they do not normally consume, a gift or coupon for a product they planned to purchase, a digital reward (e.g., customization items for a user avatar, achievement badges, unlocked content, in-game currency, etc.), etc.

108 108 108 108 200 In some embodiments, the lifestyle AI coachmay assess baseline patterns of glycemia, including the frequency and timing of glucose checks and the resulting glucose levels. For example, if a user consistently checks fasting blood sugar levels, which are generally within normal range, the AI coachmay prompt the user to also check glucose levels 1-2 hours after meals. This additional data may reveal postprandial glucose spikes, which could indicate impaired glucose tolerance and potential prediabetes presentation. Upon detecting such patterns, the lifestyle AI coachmay suggest specific dietary modifications. For example, the AI coachmay recommend replacing a pastry and coffee breakfast with steel-cut oatmeal and nuts, or advise moving dinner to an earlier time. The systemmay then track the impact of these changes, potentially observing a reduction in glucose excursions.

108 108 200 108 In some embodiments, the lifestyle AI coachmay be programmed to address common challenges in both short-term and long-term behavior change. For short-term challenges, the AI coachmay help users overcome inertia by suggesting small, manageable lifestyle changes and demonstrating their immediate benefits. This approach may prevent users from becoming overwhelmed by attempting to implement multiple significant changes simultaneously. For long-term challenges, the systemmay be designed to combat issues such as routine fatigue, plateaus in progress, holiday-related setbacks, and/or health challenges. For example, the lifestyle AI coachmay provide varied recommendations to maintain engagement, offer strategies for overcoming plateaus, and provide tailored guidance during holidays or periods of health challenges.

200 108 110 108 200 108 The systemmay also address an importance of social support in maintaining behavior changes. For example, the lifestyle AI coachmay inquire about trusted family members or friends (e.g., buddy devices) that may provide support and accountability. With user permission, the AI coachmay share guidelines and incentives with these trusted individuals, helping to understand how to provide support to benefit the user. Additionally, the systemmay use web-searching functions to identify activities or locations of interest to both the user and their support network, potentially aligning healthy choices with social activities. In some embodiments, the lifestyle AI coachmay directly contact the user's support network to let them know of the user's status and/or to provide scheduled check-ins to further incentivize the user to keep engaged and on track with guidance provided.

108 108 In some embodiments, the lifestyle AI coachmay incorporate elements of cognitive behavioral therapy, focusing on self-worth and positive self-perception. This approach may help users avoid viewing themselves as impaired or diseased, which could lead to self-fulfilling negative expectations. Instead, the AI coachmay emphasize positive reinforcement, particularly when data shows immediate improvements in health metrics.

200 108 In some embodiments, the systemmay address specific hurdles that users commonly face when attempting to adopt evidence-based dietary guidelines. These may include rigid work and home schedules, family resistance to food changes, taste adaptation challenges, and social expectations around food habits. The lifestyle AI coachmay provide targeted strategies for each of these challenges, such as suggesting gradual reductions in salt and sugar intake to allow for taste adaptation, or offering ideas for maintaining healthy choices within existing social routines.

108 200 108 The lifestyle AI coachmay be programmed to maintain motivation even after users achieve initial health goals or start medication regimens. By providing updated recommendations and/or nudges, the systemmay continue emphasizing the ongoing benefits of healthy behaviors and the potential to prevent future complications. In this way, the AI coachmay help users avoid the common pitfall of relaxing healthy habits after reaching a perceived endpoint.

200 108 In some embodiments, the systemmay provide continued support during periods of high stress, vacations, holidays, or illness, when users are more likely to deviate from healthy behaviors. For example, the lifestyle AI coachmay offer modified recommendations during these periods, helping users maintain some degree of healthy habits without adding undue stress.

200 17 18 FIGS.A-B Through the varied and adaptive approaches described herein, the systemmay provide comprehensive, personalized support for users seeking to improve health behaviors and management of glucose levels effectively. Additional examples are described in.

3 FIG. 2 FIG. 300 300 102 300 304 302 306 302 308 308 228 310 104 312 302 illustrates a block diagram of an example end to end systemfor generating noninvasive glucose assessments associated with a body site of a user. The systemmay include the wearable device() placed substantially adjacent to a body site of the user, such as a volar wrist. In operation, the systemmay illuminate () a body site(e.g., a hand or other body part) having one or more visible venous blood vessels. The illuminating may cause () reflected light to scatter and absorb at the body site. The scattered light may be captured () by an image sensor(e.g., image sensor, camera, optical sensor, etc.). The captured image may be processed () by one or more deep learning models (e.g., ML models) to generate () a glucose level prediction for the user associated with the body site.

300 300 308 102 300 102 200 300 2 FIG. In general, the systemmay capture images of veins at the body site. During illumination of the body site, changes in glucose can be monitored and compared over time using system, as described in detail in. Changes in the concentration of glucose may modify the indexes of refraction of both the intracellular medium and the extracellular medium, which in turn alters the scattering properties of those media. Such changes in scattering and photon absorption in clinical measurements, as described in detail in Table 1 below, suggest a strong correlation between a reduced scattering coefficient and the concentration of glucose as determined from a blood test. As will be described in detail herein, the image recorded by an NIR image sensordepicts a correlation between glucose level (as read by over-the-counter (OTC) glucose meter) and the darkness/appearance of veins and the surrounding area. The system executing on deviceand/or systemmay be used to detect particular glucose measurements and/or determine glucose-based or health based predictions about a user wearing devicebased at least in part on this correlation. Thus, upon capturing one or more images of a body site, the one or more images may be stored and programmatically analyzed (i.e., using a computer system using hardware and software) to obtain a predicted output value of glucose for the user. The systems,may learn to recognize any change in the absorption and/or scattering of NIR light within a vein to ascertain an estimated change in a glucose level in the blood of the user.

4 4 FIGS.A-B 4 4 FIGS.A-B 400 400 402 402 102 402 228 102 102 102 illustrate photographic imagesA,B depicting IR light reflecting off of an arm (e.g., volar wrist) of a user. In this example, the volar wristmay represent a body site on which devicemay be worn. Alternatively, the volar wristmay represent a body site that is placed into contact with an image sensor (e.g., image sensor), detector, or other element of deviceworn on another body part in order to generate a glucose prediction/measurement for the user. While the images shown inmay depict a swatch of the body site that is larger than what may be captured and/or analyzed by device, the images represent at least portions of the arm that could be captured and analyzed by the device. The systems and methods described herein may of course, capture and analyze less than an entire image of a particular body part while maintaining the ability to ascertain blood glucose levels associated with less than the entire body site, and/or the ability to ascertain other physiological parameters or conditions.

4 FIG.A 400 102 400 228 102 102 illustrates a photographic imageA of a user with a glucose meter reading of about 93 mg/dL. Here, the wearable devicemay capture at least one digital imageA representing infrared light reflected from the arm. Here, the arm is a non-limiting example of a body site in an upper extremity of the user such that a sensor (e.g., image sensor) installed on the wearable device, for example, may capture at least one digital image of the volar wrist of the user when the wearable deviceis worn on the wrist.

102 102 228 102 102 400 104 400 404 406 408 410 104 400 a a a a a a The wearable device(and/or other computing device communicatively coupled to device) may programmatically analyze the captured digital image(s) to ascertain that digital data captured by the image sensorcorresponds to or is strongly suggestive of particular glucose meter reading values based on the characteristics of the reflected light. For example, the wearable device(and/or other computing device communicatively coupled to device) may programmatically analyze the imageA to assess blood glucose for a user. The analyzing may include segmenting, by a first ML model (e.g., ML model) the imageA into a plurality of regions. The regions may each include at least a portion of a venous blood vessel (e.g., vessel, vessel, vessel, and/or). The segmenting performed by ML modelmay be performed to detect at least one venous blood vessel (or at least one portion of the venous blood vessel) in the imageA, for example.

102 102 104 206 104 208 210 b b The analyzing performed by wearable device(and/or other computing device communicatively coupled to device) may further include analyzing, by a second ML model (e.g., ML model), a visual appearance of at least one of the identified venous blood vessels in the plurality of regions of the image. Analyzing a visual appearance may include analyzing a pixel intensity (e.g., darkness, brightness, etc.) of each venous blood vessel or portion of such a vessel represented in a given image. Each intensity level determined for a region in an image may be compared to predefined glucose data and/or patient datato determine an accurate glucose level and/or glucose prediction for the user. The ML modelmay use mapping logicand/or prediction systemto determine the accurate glucose level and/or glucose prediction for the user in near real time.

400 404 406 508 410 a a a a For example, analyzing the visual appearance of one or more venous blood vessels in the plurality of regions includes determining an intensity level of pixels representing a portion of the venous blood vessel(s). For instance, the intensity level of pixels may represent a darkness level or brightness level that may be ascertained by processing the imageA and determining that pixels representing the vessel portionare at a first intensity level, while vessel portions,, andare at different intensity levels. The intensity levels may be further analyzed in combination and/or with other data to generate a glucose level and/or glucose prediction for the user.

400 206 104 404 406 408 410 412 104 400 404 410 102 206 b a a a a b a a For example, additional analysis of the veins in the imageA may be performed and combined with other analysis (e.g., determined intensity levels of particular veins, prior glucose predictions or measurements, or the like) and/or patient data. In some embodiments, the ML modelmay also determine a dehydration level and/or an osmolality level based on one or more features of the pixels representing the portion of at least one venous blood vessel (e.g., vessel, vessel, vessel, and/or). Such features may include a blood vessel diameterof the portion of one or more venous blood vessel in the image. For example, the ML modelmay use the imageA to determine a vessel diameter of any portion of vessels-, for example. The determination of dehydration levels and/or osmolality levels may be compared to population level data, prior images associated with the user wearing device, and/or patient datain order to determine changes in vessel diameter, vessel behavior, and/or vessel size, in general.

104 400 104 400 b b Another feature that the ML modelcan determine is a volume of fluid in the portion of the one or more venous blood vessels in the imageA. For example, the ML modelmay analyze the fluid, particulates, and/or cross sectional areas and any associated volumes pertaining to the one or more venous blood vessels in the imageA.

104 400 400 b In addition, the second ML modelmay also perform a classification process to classify the plurality of regions according to predefined glucose levels. For example, each of the veins in the imageA may be classified into a particular window of glucose measurements based on a visual appearance in an image including any or all of the determined pixel intensity levels, the determined dehydration level approximations, the determined osmolality level approximations, the determined vessel diameters and related size-based parameters, and/or the determined volumes associated with at least a portion of one or more of the venous blood vessels depicted in imageA, for example.

102 102 102 102 250 102 102 The wearable device(and/or other computing device communicatively coupled to device) may then generate, based on the classifying and on the visual appearance of the at least one venous blood vessel, a blood glucose estimate for the user. That is, the wearable device(and/or other computing device communicatively coupled to device) may utilize the pixel intensity levels, dehydration level approximations, osmolality level approximations, vessel diameters and related size-based parameters, and volumes associated with at least a portion of one or more venous blood vessels to generate a predicted glucose level and/or measurement for a user of the one or more venous blood vessels. Additional indicationsmay also be provided to a user using device(and/or other computing device communicatively coupled to device) based on the generated predicated glucose level and/or measurement.

4 FIG.B 4 FIG.A 4 FIG.A 4 FIG.A 400 400 404 404 406 406 408 408 410 410 404 406 408 410 4 404 406 408 410 a b a b a b a b b b b b a a a a illustrates a photographic imageB of the same user ofwith a glucose meter reading of about 89 mg/dL. The imageB was captured after the user ofingested a meal. The glucose level has dropped about 5 points from 93 mg/dl to 89 mg/dL. Visual changes are apparent between vesselsand,and,and, and between vesselsand. Namely, each vessel,,, andof FIG.B has less intensity, less volume, and less size than that of the veins,,, andof.

102 102 404 406 408 410 404 406 408 410 104 4 FIG.A 13 FIG. 4 FIG.B 4 FIG.B b b b b b b b b b The wearable device(and/or other computing device communicatively coupled to device) may perform the same process as described in(and/or) to ascertain parameters including, but not limited to the pixel intensity levels, dehydration level approximations, osmolality level approximations, vessel diameters and/or related size-based parameters, or volumes associated with at least a portion of vessels,,, andshown in.represents the vessels,,, andat a time after a meal event. The parameters may be used to determine a corresponding glucose level for the user at the time after the meal event. Classification or reclassification of outputs from ML modelmay also be performed to determine an updated predicted output value (e.g., glucose level) for the user.

5 5 FIGS.A-B 500 500 502 102 illustrate photographic imagesA,B depicting IR light reflecting off of an arm (e.g., volar wrist body site) of a user before and after an event, respectively. The event described here includes ingestion of a meal, but any event that may alter blood glucose may be substituted and a blood glucose prediction/measurement can be made by wearable deviceand/or other computing system described herein. Table 1 below indicates the event/meal and the changes in glucose levels, as measured by a glucose meter.

502 102 502 228 102 102 102 5 5 FIGS.A-B In this example, the body sitemay represent a body site on which wearable devicemay be worn. Alternatively, the body sitemay represent a body site that is placed into contact with an image sensor (e.g., image sensor), detector, or other element of deviceworn on another body part in order to generate a glucose prediction/measurement for the user. While the images shown inmay depict a swatch of the body site that is larger or smaller than what may be captured and/or analyzed by device, the images represent at least portions of the arm that could be captured and analyzed by the device. The systems and methods described herein may, of course, capture and analyze less than an entire image of a particular body site while maintaining the ability to ascertain blood glucose levels associated with less than the entire body site, and/or the ability to ascertain other physiological parameters or conditions.

5 FIG.A 500 102 102 502 102 504 506 508 102 102 504 506 508 102 504 506 508 504 506 508 500 a a a a a a a a a a a a illustrates a photographic imageA of a user with a glucose meter reading of about 97 mg/dL. The wearable device(and/or other computing device communicatively coupled to device) may analyze the body siteto determine and/or predict a blood glucose level for the user wearing deviceat a particular point in time, as described in detail herein. In this example, the analysis may be based at least in part on visually assessing a venous blood vessel, a venous blood vessel, and/or a venous blood vessel. For example, the wearable devicemay analyze a visual appearance of one or more vessels to determine a blood glucose level for the user at a point in time. For example, the wearable devicemay analyze the visual appearance of vessel,, and/orto determine a blood glucose level for the user wearing device. Analyzing the visual appearance of vessels,, and/ormay include, but is not limited to, assessing pixel intensity, osmolality, blood vessel diameter, blood vessel size, vessel length, vessel width, vessel cross-sectional area, or the like of a portion of one or more venous blood vessels,, and/orin the imageA.

5 FIG.B 500 102 500 102 502 500 504 506 508 504 506 508 504 506 508 504 506 508 504 506 508 500 500 102 b b b b b b a a a b b b a a a illustrates a photographic imageB of a user with a glucose meter reading of about 92 mg/dL. In this example, the user wearing deviceingested a meal and the imageB depicts a reading directly after ingestion of the meal. The wearable devicemay assess the visual appearance of body siteand in particular, of the same vessels, shown in imageB as vessels,, and. The vessels,, andare depicted with greater pixel intensity than vessels,, and. In addition, vessels,, andare larger (e.g., wider and longer) than vessels,, and. However, the changes between imageA and imageB may not be apparent to the visual eye to provide a blood glucose level or measurement. As such, the wearable deviceand in general the systems described herein may assess the images to determine and/or predict specific blood glucose levels and/or measurements such that the user may make informed decisions to maintain and/or improve blood sugar health, or the like.

6 6 FIGS.A-B 6 6 FIGS.A andB 5 5 FIGS.A andB 102 500 1 102 illustrate photographic images of a volar wrist with mapped veins before and after an event, respectively. The user depicted inis the same user and timing as shown in, respectively. Here, the wearable deviceand/or the systems described herein may analyze the visual appearance of venous blood vessels (and/or vessel portions) depicted in the imageAto determine a first glucose measurement and/or prediction. The device, for example, may map particular vessels with specific measurements, overlays, and/or coordinates that may be used for future analysis of glucose for the user at a later time. For example, at a later time after an event, such as a meal.

6 FIG.B 6 FIG.A 102 500 1 602 602 a b illustrates the same user ofafter a meal event. The wearable devicemay next analyze the visual appearance of the same venous blood vessels (and/or vessel portions) shown in imageBto generate a second glucose measurement and/or prediction, which may also be marked and/or mapped with measurements, overlays (e.g., overlay,, etc.), and/or coordinates corresponding to changes in the visual appearance of one or more vessels.

602 500 1 602 500 1 602 104 a b a In some embodiments, the first glucose measurement can be compared to the second glucose measurement to generate indications for the user, as described in detail herein. For example, the overlayin imageAdepicts a vessel as a particular width and length while overlayin imageBdiffers from the width and length of overlay. The differences can be analyzed by ML modelsto identify specific changes in the vessel(s) over time to determine occurrences of blood glucose changes and estimates of blood glucose levels. Any number of glucose measurements and/or vessel analysis may be performed to generate additional indications or related data.

7 7 FIGS.A-B 7 FIG.A 7 FIG.B 702 702 102 702 illustrate photographic images of a volar wrist body sitecaptured before and after a meal, respectively.illustrates the body sitebefore a meal whileillustrates the body site after a meal, as indicated in Table 1 below. In this example, the wearable devicemay again analyze the change in visual appearance of vessels within body siteto ascertain changes in glucose over time, as described in detail herein.

8 8 FIGS.A-B 8 FIG.A 8 FIG.B 8 FIG.B 802 802 802 102 802 802 102 802 802 102 a b a b a b illustrate photographic images of a volar wrist body siteand body sitecaptured before and after a meal, respectively.illustrates the body sitebefore a meal whileillustrates the body site after a meal, as indicated in Table 1 below. In this example, the wearable devicemay again analyze the change in visual appearance of vessels within a first body siteto ascertain changes in glucose over time by comparing to the same vessels or different vessels in a second and different body site. For example, if wearable deviceshifts on the body siteto body site, assessments and predictions may still be performed by deviceby assessing any image overlap and/or assessing additional vessels in the latest image in time (e.g.,).

9 9 FIGS.A-B 9 FIG.A 9 FIG.B 902 102 902 illustrate photographic images of a volar wrist captured before and after a meal, respectively.illustrates the body sitebefore a meal whileillustrates the body site after a meal, as indicated in Table 1 below. In this example, the wearable devicemay again analyze the change in visual appearance of vessels within body siteto ascertain changes in glucose over time, as described in detail herein.

904 904 904 904 102 102 a b a b 9 9 FIGS.A andB 9 FIG.B 9 FIG.A 9 FIG.B 9 FIG.A 9 FIG.B A vesselis shown in both. In, the vesselis shown to have grown larger than vesselshown in. In particular, in, the vesselis shown to be bulging. This change may be detected by wearable deviceand further analysis can be performed to identify specific changes in blood glucose that occurred between the image captures ofand. In this example, the body site is shown as a circular shape, however, any shape or size of a capture area on a body may be used by wearable device.

10 10 FIGS.A-B 10 FIG.A 10 FIG.B 1002 1002 102 1002 illustrate photographic images of a volar wrist body sitecaptured before and after a meal, respectively.illustrates the body sitebefore a meal whileillustrates the body site after a meal, as indicated in Table 1 below. In this example, the wearable devicemay again analyze the change in visual appearance of vessels within body siteto ascertain changes in glucose over time, as described in detail herein.

11 11 FIGS.A-B 11 FIG.A 11 FIG.B 1102 102 1102 illustrate photographic images of a volar wrist captured before and after a meal, respectively.illustrates the body sitebefore a meal whileillustrates the body site after a meal, as indicated in Table 1 below. In this example, the wearable devicemay again analyze the change in visual appearance of vessels within body siteto ascertain changes in glucose over time, as described in detail herein.

12 12 FIGS.A-D 12 12 FIGS.A-D 12 12 FIGS.A-D 12 12 FIGS.A-D 12 12 FIGS.A-D 102 illustrate example placement of an example noninvasive glucose monitoring device. The noninvasive glucose monitoring devices described herein may take any form or shape that may be fitted to a body site in which one or more venous blood vessels are visible under the skin associated with the body site. The devices ofmay represent wearable device. The devices ofare depicted on or adjacent to a dextral or volar wrist of a user. Although the devices ofare depicted as a smart watch, other form factors are possible including one or more of a device integrated into a band, a watch, an adhesive strip, jewelry, a patch, clothing, or any other wearable accessory. Various shapes of the devices ofare depicted. However, other shapes are possible.

12 FIG.A 1200 1202 228 1200 1200 1202 1200 illustrates an example wearable deviceon a volar wristof a user. In this example, an image sensor (e.g., image sensor) may be coupled to (or integrated into) a base portion (not shown) of device. In operation, the base portion is placed in contact with a portion of the volar wrist. The devicemay capture one or more images of the portion of the volar wristof the user wearing device.

12 FIG.B 12 12 FIGS.C andD 1200 1204 228 1206 1200 1202 1208 1200 1208 1200 1200 illustrates the example wearable deviceon a dextral wristof a user. In this example, an image sensor (e.g., image sensor) may be coupled to (or integrated into) a strap portionof the deviceto capture images of a portion of the volar wristof the user. In some embodiments, the image sensor may instead be coupled to (or integrated into) a top portionof device. In such an arrangement, images may be captured of one or more venous blood vessels in response to a user placing a body site in contact or near contact (e.g., hover) with the top portionand devicedetecting such contact or near contact, as shown in. In addition, such an arrangement may also allow for deviceto be placed in contact (or near contact) with another body portion including, but not limited to a neck, a lymph node region, or any portion of a limb, an abdominal region, or a thoracic region.

12 FIG.C 12 12 FIGS.C-D 1210 1210 228 1210 1210 1212 1212 1212 1214 1210 1216 1214 1210 1214 1210 1210 illustrates an example wearable devicefor capturing images of body sites that are not wearing the wearable device. For example, devicemay include one or more image sensors (e.g., image sensors) on any surface of the device facing away from the skin in contact with the device. Such surfaces may include such sensors and illumination devices to capture images of body sites when the devicedetects a nearness or contact with a particular body site. For example, the devicemay be worn on a sinistral dorsal wristwith a sensor and/or light source positioned on the sinistral dorsal wristand facing away from the sinistral dorsal wrist. The user may move the dextral volar wristtoward the face of device, as shown by arrow. When the sensor detects that the dextral volar wristis within a predefined range of the light source on device, at least one digital image captures at least a portion of the dextral volar wrist. Althoughshow the deviceon a sinistral wrist and detection occurring on the dextral wrist, one of skill in the art will appreciate that the devicemay be worn on the dextral wrist and the detection may occur on the sinistral wrist.

12 FIG.D 12 FIG.C 1210 1210 1210 illustrates the example wearable deviceofcapturing images of a body site that is not wearing the wearable device. In this example, the devicemay capture images of a body site that is placed on (or adjacent to) a sensor on device.

13 FIG. 1300 1300 102 is a flow diagram of an example processfor performing noninvasive blood glucose monitoring of a user. The processmay be performed to generate one or more predictions or measurements of a blood glucose level in a user wearing a monitoring devices, such as device.

1300 218 1300 104 210 208 104 210 208 230 228 204 1404 202 102 102 102 1300 102 204 The processmay be executed on a processor (e.g., processors). The processmay utilize ML models, prediction system, and mapping logic. Each of the ML models, prediction system, and mapping logicmay be in communication with one or more sensor interfacesand sensors, apps, processors, and/or memoryof deviceto receive and process sensor signals captured at a body site of a user. The sensor signal may be obtained using deviceor another device communicatively coupled to device. The processmay be performed on device, on a companion device (e.g., mobile device), in an on-device app (e.g., app), in a third-party app, or on the cloud, or the like.

1302 1300 232 228 102 102 At block, the processincludes causing a light source to illuminate a body site of a subject. For example, the optional light sourceor a light source of image sensormay receive instructions to illuminate the body site of a user. The instructions may be provided by a program executing on wearable deviceor executing on a device communicatively coupled to device. The illumination may provide NIR light to the body site.

1304 1300 228 228 102 At block, the processincludes capturing at least one digital image representing reflected light from the body site. For example, the image sensormay capture one or more images (e.g., image signals, digital images, light reflections, digital video, one or more sequences of image frames, etc.) of one or more venous blood vessels (or venous blood vessel portions) viewable on the surface of skin at the body site. In operation, the image sensormay be installed on devicein an arrangement to capture an image/image signal of at least one venous blood vessel portion in the image capture. The portion can be used to perform the analysis of glucose levels for the user.

1306 1300 102 102 102 102 At block, the processincludes programmatically analyzing the at least one image to assess blood glucose for the subject. For example, the wearable device(and/or other computing device communicatively coupled to device) may programmatically analyze the captured image(s) to determine a blood glucose level or measurement for the user based at least in part on the characteristics of the reflected light. For example, the wearable device(and/or other computing device communicatively coupled to device) may programmatically analyze a captured image to assess blood glucose levels.

102 In some embodiments, programmatically analyzing the at least one digital image may be performed to compare the at least one digital image to any number of previously captured prior digital images of the body site as a way to ascertain changes in blood glucose levels of the user in response to an event. For example, the analysis of digital images captured by devicecan be assessed and compared over time to determine past behavior of glucose levels for a specific user and infer future behavior of the glucose levels for the specific user.

1308 1300 250 At block, the processincludes generating an indication (e.g., indications) representing the blood glucose estimate. For example, the analysis described herein may result in an output that is a blood glucose estimate and/or related prediction representing a glucose level for the user of the body site. The output may be further analyzed to generate an indication that may be communicated to an app, the user, or other user with permissions to receive glucose data about the user under test. The indication may be textual, audial, visual (e.g., messages, notifiers, animations, graphical content, image based content, etc.), or a combination thereof, as described in further detail herein.

In some embodiments, the blood glucose estimate may be expressed as a (a) range of values or as (b) a category in a plurality of categories. For example, each class of image features/glucose levels may include a class size of about 5 mg/dL. A total range of clinical significance may be about 54 mg/dL to about 250 mg/dL. In such an example, the outputs of the ML classifier may include (i) about 54 mg/dL to about 59 mg/dL, (ii) about 60 mg/dL to about 65 mg/dL (iii) about 66 mg/dL to about 71 mg/dL, etc. The total number of such output classes may be about 33 classes. In the event that class sizes are configured to be greater than 5 mg/dL, fewer total output classes may be used. Modifying the class size may be used to increase the accuracy of the ML models described herein.

In some embodiments, a classification is not performed and instead a machine learning regression model is utilized by the ML models described herein. In such an example, each integer value in the range of about 54 mg/dL to about 250 mg/dL may represent a potential output of the machine learning regression model.

In some embodiments, the classification may include classifying image features/glucose levels according to color coding. The color coding may pertain to ranges as described elsewhere herein. However, this color coding may ensure that a user may understand a glucose level by recognizing colors that familiarly indicate low glucose, high glucose, average glucose, etc. For example, green may indicate that blood glucose levels are average or good such that no action is being asked of the user. Orange or red may indicate that glucose is becoming higher than average (e.g., orange) or becoming dangerously high or low (e.g., red), which may signal to the user to perform particular actions to correct the glucose level.

In some embodiments, the classification may include classifying image features/glucose levels according to an application-specific unit or scale. For example, the classification may include classifying a particular subject into a particular glycemic level over a predefined number of hours or days.

In some embodiments, the classification may include outputting a change in glucose value. Such a classification may avoid calibration steps that may otherwise be used. For example, a change in glucose after exercise may be −5 mg/dL. Further, for example, a change in glucose after sleeping (e.g., a fasting state) may be −15 mg/dL.

204 250 In some embodiments, the indication representing the blood glucose estimate and/or related prediction may include application data (e.g., for or from one or more app(s)). The application data may be arranged in a way to present the blood glucose measurements and/or related predictions to the user in one or more of: an email, a display screen, a text message, an audio entry, and a visual indicator (e.g., indicator(s)) within a computing device or wearable device associated with the subject or a care provider of the subject.

In some embodiments, the indication representing the blood glucose estimate and/or related prediction may be provided as a notification to one or more interested parties who have (a) indicated an interest in receiving notifications and (b) are approved by the subject. For example, the indication may be associated with one or more rules for sharing indications about a particular user in which the indication pertains.

250 204 In some embodiments, a predicted output value or indicationabout a glucose level for a user may be expressed as a specific number or a range or category. The predicted output value may also be expressed as a measurement or estimate that is estimated to be accurate within a predefined margin of error. The predicted output value may be, at least in part, expressed using one or more colors and/or audio cues within an app, for example, to help inform a user, particularly in ways that can reach audiences of varying ages (e.g., children, the elderly, care providers, etc.) and abilities (e.g., hearing or visually impaired). For example, a certain color and/or audio cue may correspond with a certain category or level of glucose in the blood. In this way, if a user has a difficult time interpreting a numerical value, the use of color on the display and/or audio cue may assist in informing the user about a determined glucose level or prediction.

1306 104 404 406 408 410 104 a a a a a a 4 FIG.A With respect to block, in some embodiments, the analyzing may include segmenting, by a first ML model (e.g., ML model) one or more images into a plurality of regions. The regions may each include at least a portion of a venous blood vessel (e.g., vessel, vessel, vessel, and/orof). The segmenting performed by ML modelmay be performed to detect at least one venous blood vessel (or at least a portion of one or more venous blood vessels).

102 102 104 206 104 208 210 b b The analyzing performed by wearable device(and/or other computing device communicatively coupled to device) may further include analyzing, by a second ML model (e.g., ML model), a visual appearance of at least one of the identified venous blood vessels in the plurality of regions of the image. Analyzing a visual appearance may include analyzing a pixel intensity (e.g., darkness, brightness, etc.) of each venous blood vessel or portion of such a vessel represented in a given image. Each intensity level determined for a region in an image may be compared to predefined glucose data and/or patient datato determine a glucose level estimate (e.g., a glucose prediction) for the user. The ML modelmay use mapping logicand/or prediction systemto determine the glucose level estimate for the user in near real time.

400 For example, analyzing the visual appearance of one or more venous blood vessels in the plurality of regions includes determining an intensity level of pixels representing a portion of the venous blood vessel(s). For instance, the intensity level of pixels may represent a darkness level or brightness level that may be ascertained by processing the imageA and determining that pixels representing a first vessel (or vessel portion) are at a first intensity level, while additional vessels (or vessel portions) are at various different intensity levels. The combination of intensity levels of each vessel and/or vessel portion may be further analyzed with other data to generate a glucose level estimate for the user.

206 104 102 206 b In some embodiments, additional analysis of the veins in one or more captured images may be performed and combined with other analysis (e.g., the determined intensity levels of particular veins, prior glucose predictions or measurements, or the like) and/or patient data. For example, the ML modelmay also determine a dehydration level and/or an osmolality level based on one or more features of the pixels representing the portion of at least one venous blood vessel. Such features may include a blood vessel diameter or size of a portion of one or more venous blood vessel in the one or more images. The determination of dehydration levels and/or osmolality levels may be compared to population level data, prior images associated with the user wearing device, and/or patient datain order to determine changes in vessel diameter, vessel behavior, and/or vessel size. In some embodiments, the vessel size may include length, width, cross-sectional area, etc.

104 102 104 b b Another feature that the ML modelcan determine is a volume of fluid in the portion of the one or more venous blood vessels in one or more image captured by device. For example, the ML modelmay analyze the fluid, particulates, and/or cross-sectional areas and any associated volumes pertaining to the one or more venous blood vessels in one or more of the captured images.

104 b In addition, the second ML modelmay also perform a classification process to classify the plurality of regions according to predefined glucose levels. For example, each of the veins or vein portions in one or more captured images may be classified into a particular window of glucose measurements based on a visual appearance in an image including any or all of the determined pixel intensity levels, the determined dehydration level approximations, the determined osmolality level approximations, the determined vessel diameters and related size-based parameters, and/or the determined volumes associated with at least a portion of one or more of the venous blood vessels depicted in the one or more captured images.

102 102 102 102 The wearable device(and/or other computing device communicatively coupled to device) may then generate, based on the classifying and on the visual appearance of the at least one vein, a blood glucose estimate for the user. That is, the wearable device(and/or other computing device communicatively coupled to device) may utilize the pixel intensity levels, dehydration level approximations, osmolality level approximations, vessel diameters and related size-based parameters, and volumes associated with at least a portion of one or more veins (or vein portions) to generate a predicted glucose level estimate (and/or related predictions or measurements) for a user of the one or more veins or vein portions.

1300 1300 102 104 106 1 FIG. In some embodiments, the processmay further include monitoring the subject over time. For example, the processmay include monitoring the subject with wearable deviceby repeating the illumination of the body site, repeating the capturing of digital images to capture additional digital images of the venous body site during the monitoring process, and performing additional programmatical analysis of the additional digital images to determine any number of additional blood glucose estimates or predictions for the user. The ML modelsmay perform the additional programmatical analysis to generate a visual mapping depicting the blood glucose estimate and the additional blood glucose estimates in a time-based graph (e.g., shown inby computing device) and as described in additional detail throughout this disclosure.

In some embodiments, the monitoring may be performed continually. In some embodiments, the monitoring may be performed intermittently. In some embodiments, the monitoring may be performed according to a predefined schedule. In some embodiments, the monitoring may be performed according to a predefined event schedule. In some embodiments, the monitoring may be performed on-demand.

1300 104 1300 104 104 b b b In some embodiments, the processmay further include generating a first set of measurements associated with features of the at least one venous blood vessel detected in the at least one digital image at a first point in time. For example, the ML modelmay assess segmented images of a first set of measurements associated with veins or vein portions in the at least one digital image at time (t). Next, the processmay include generating a second set of measurements for at least one image of the additional digital images corresponding to the same body site in the at least one digital image. The second set of measurements may correspond to the features of the at least one venous blood vessel at a second point in time. Next, the ML modelmay compare the first set of measurements to the second set of measurements to determine an updated glucose measurement for the user. Additional assessments and/or predictions may also be performed by ML modeland such assessments and/or predictions may be provided as output as described in detail throughout this disclosure.

1300 102 1300 In some embodiments, the processmay be carried out on a computing device in communication with the wearable device. For example, the processmay be executed on one or more processors of a smart phone, a smart watch, a mobile gaming device, a laptop computer, a desktop computer, a cloud hosted computer, or the like.

1300 1300 In such an arrangement, the methodmay include performing noninvasive blood glucose monitoring of a user. The process may include causing a light source to illuminate a body site of the user, capturing at least one digital image representing reflected light from the body site, programmatically analyzing the at least one digital image by the at least one processor to assess blood glucose for the user. The analyzing may include receiving a segmentation output from a first machine learning model. The segmentation output may include identified venous blood vessel portions in the at least one digital image and image segments representing a plurality of regions of the at least one digital image. The analyzing may further include receiving analysis output from a second machine learning model. The analysis output may include identification of one or more portions in the segments of the at least one digital image and a visual appearance assessment of the portions of the venous blood vessel in the plurality of regions. The analysis may further include receiving, based on the segmentation output and the analysis output, a blood glucose estimate for the user. The processmay include generating an indication representing the blood glucose estimate.

104 102 1300 In some embodiments, the first machine learning model and the second machine learning model are accessed on a cloud-based computing device (e.g., cloud servers executing ML models) by a noninvasive blood glucose monitor (e.g., wearable device) that may monitor the user. For example, processmay be performed on a cloud-based computing device.

In some embodiments, the visual appearance assessment of the portions of the venous blood vessel in the plurality of regions includes determining an intensity level of pixels representing a portion of the at least one venous blood vessel in the at least one digital image and determining an osmolality level based on one or more features of the pixels representing the portion of the at least one venous blood vessel in the at least one digital image. Such features may include a determined vessel diameter of the portion of the at least one venous blood vessel in the at least one digital image and/or a determined volume of fluid in the portion of the at least one venous blood vessel in the at least one digital image.

1300 In some embodiments, the processmay include monitoring, the subject over time, to capture a plurality of additional images, programmatically analyzing the additional images to determine a plurality of additional blood glucose estimates for the subject, and generating a visual mapping depicting the blood glucose estimate and the plurality of additional blood glucose estimates in a time-based graph.

228 228 228 In some embodiments, the image sensormay exchange data (such as captured image and video) over a wireless network or Bluetooth connection to a computer system designed to perform analysis on the image (and potentially video) data captured by the image sensor. In some embodiments, that data may be exchanged over a physical link; for example, the software responsible for performing analysis on the captured digital images/videos may reside and execute in the same physical structure as the image sensor.

232 228 1400 106 102 In some embodiments, the light source, image sensor, and computer system (e.g., computer systemor computing device) for running image analysis may be designed with a convenient form factor adaptable for use in a convenient location, such as the kitchen, car, bathroom, or office. Embodiments may also be incorporated into wearable technology, such as wearable device, or another device that may be incorporated into clothing or otherwise worn/carried by a human. Embodiments may also be incorporated into mobile phones and other portable devices.

In embodiments where the accuracy of the glucose readings has exceeded a specified threshold, then an alert can be issued to the user and potential other interested/approved parties (such as a care provider or family member) when a glucose level exceeds a predefined threshold of a predefined normal range for the user.

Embodiments may be used to provide useful information to a user pertaining to user health and wellness on a range of topics, including but not limited to (a) the location of veins in the body, (b) the size, diameter, and/or width of veins in the body, (c) the propensity of change in glucose levels after particular predefined or detected events, such as one or more of: a meal, a fast, a physical stressor, an emotional stressor, a sleep event (e.g., excess sleep, sleep deprivation, exhaustion, etc.), a time of day, and a change in estrogen (d) an indirect measurement of dehydration and/or osmolality by measuring water content in the blood or directly measuring osmolality, (e) measuring the fat content in the blood carried by the veins, and/or (f) information about any substance in blood that shows a specific spectral signature in the wavelength range from about 700 nm to about 2500 nm.

5 FIG.A 5 FIG.B 7 FIG.A 7 FIG.B 8 FIG.A 8 FIG.B 9 FIG.A 9 FIG.B 10 FIG.A 10 FIG.B 11 FIG.A 11 FIG.B Table 1 below includes data results from an experiment that compared readings generated by a glucometer and readings generated by image sensors. In the experiment, pre-meal and post-meal readings were taken from 33 non-diabetic individuals with consent. Post-meal readings from all participants were taken at exactly 1.5 hours after commencement of a meal. In this example, five different optical sensors were evaluated including Shortwave infrared (SWIR) cameras used with different light sources in the about 1050 nm to about 1550 nm. The Abbott FREESTYLE LITE finger prick glucose meter was used as a reference. The total number of data points collected was 408. Images associated with the experiment are shown in,,,,,,,,,,, and.

TABLE 1 Data Results Before After meal meal gluco- gluco- meter meter User Age Race/ reading reading Delta (NIR) #ID Range Gender Ethnicity (mg/dL) (mg/dL) mg/dL Meal outputs  1 40-44 M Indian 82 132 50 Indian white FIG. naan and 7A, 7B paneer (Indian cheese) 11 40-44 M Chinese 98 110 12 unknown FIG. 5A, 5B 12 55-59 F Vietnamese 96  96  0 corn chips, FIG. cheese pizza, 8A, 8B salsa, a little cabernet sauvignon 21 40-44 F Indian 97  92 −5 2 slices of FIG. veggie pizza 10A, 10B and a cup of Indian chai 26 50-54 M Indian 71 121 50 3 wheat rotis, FIG. 3/4 cup okra, 9A, 9B potato curry, 1/2 cup mango pulp, 1 cup spinach, cucumber, bell pepper salad (no dressing). 33 70-74 F Indian 91 108 17 Wheat roti, FIG. curry (greens, 11A, 11B cauliflower), nuts, cream

Data proved that the NIR optical sensor technology is independent of gender, race, and age. It proved about 56% correlation between the delta in finger prick reading and the observed changes by the embodiments described herein. The described embodiments may provide an advantage of detecting possible errors in the conventional glucose meter.

14 FIG. 1400 232 228 202 1400 102 is a block diagram of an example computing system for performing noninvasive blood glucose monitoring of a user and monitoring well-being and/or glucose. The computer systemwhich may control the operation of one or more of a light source (e.g., an IR light source, red-light source, light source), a digital camera (e.g., image sensor), and a software (e.g., within memory) for programmatically analyzing images/reflected light captured by the digital camera to capture or generate blood glucose estimates, predictions, and associated data. In some embodiments, systemmay represent a portion of the hardware of the wearable device.

1400 1404 218 1406 202 1408 1410 1418 1404 1406 1404 1406 1404 1400 1408 1404 1410 In some embodiments, computer systemincludes one or more processors(e.g., processors), main memory(e.g., memory), ROM, storage device, and communication interface. The at least one processormay process information. Memory, such as a random-access memory (RAM) or other dynamic storage device, may function to store information and instructions to be executed by processors. Main memoryalso may be used for storing temporary variables or other intermediate information during execution of instructions to be executed by processor. Computer systemfurther includes a read only memory (ROM)or other static storage device for storing static information and instructions for processor. A storage device, such as a magnetic disk or optical disk, is provided for storing information and instructions.

1400 1400 1404 1406 1406 1410 1406 1404 The described embodiments are related to the use of computer systemfor implementing the techniques described herein. In some embodiments, such techniques are performed by computer systemin response to processorexecuting one or more sequences of one or more instructions contained in main memory. Such instructions may be read into main memoryfrom another machine-readable medium, such as storage device. Execution of the sequences of instructions contained in main memorycauses processorto perform the process steps described herein. In some embodiments, hard-wired circuitry may be used in place of or in combination with software instructions to implement particular embodiments. Thus, the described embodiments are not limited to any specific combination of hardware circuitry and software.

1404 The term “non-transitory computer-readable storage medium” as used herein refers to any tangible medium that participates in storing instructions which may be provided to processorfor execution. Non-limiting, illustrative examples of non-transitory machine-readable media include, for example, a solid-state device, a floppy disk, a flexible disk, hard disk, magnetic tape, or any other magnetic medium, a CD-ROM, any other optical medium, a RAM, a PROM, and EPROM, a FLASH-EPROM, any other memory chip or cartridge, or any other medium from which a computer can read.

1404 1420 1400 Various forms of non-transitory computer-readable media may be involved in carrying one or more sequences of one or more instructions to processorfor execution. For example, the instructions may initially be carried on a magnetic disk of a remote computer. The remote computer can load the instructions into its dynamic memory and send the instructions over a network linkto computer system.

1418 1420 1418 1418 1418 Communication interfaceprovides a two-way data communication coupling to a network linkthat is connected to a local network. For example, communication interfacemay be an integrated services digital network (ISDN) card or a modem to provide a data communication connection to a corresponding type of telephone line. As another example, communication interfacemay be a local area network (LAN) card to provide a data communication connection to a compatible LAN. Wireless links may also be implemented. In any such implementation, communication interfacesends and receives electrical, electromagnetic, or optical signals that carry digital data streams representing various types of information.

1420 1420 Network linktypically provides data communication through one or more networks to other data devices. For example, network linkmay provide a connection through a local network to a host computer, a mobile phone, or to data equipment operated by an Internet Service Provider (ISP).

1400 1420 1418 1418 1404 1410 Computer systemcan send messages and receive data, including program code, through the network(s), network linkand communication interface. For example, a server might transmit a requested code for an application program through the Internet, a local ISP, a local network, subsequently to communication interface. The received code may be executed by processoras it is received, and/or stored in storage device, or other non-volatile storage for later execution.

15 FIG.A 1500 1500 1502 1504 1502 1504 1504 1502 1500 1502 1500 illustrates an example perspective view of a wearable devicefor monitoring glucose and user behavior over time. The wearable devicemay include a bandand an optional watch face. The bandmay be communicatively coupled to the watch faceto communicate data for display on watch face. The bandmay include electronics for monitoring glucose for a user wearing device. In addition, the bandmay include one or more processors coupled to hardware, sensors, or devices for monitoring user behavior, receiving user inputs, and/or generating/displaying recommendations for modifying such user behavior. In some implementations, the deviceis communicatively coupled to one or more other computers, smart phones, servers, ML models, or the like.

1502 1506 228 229 1508 1510 1502 1500 In this example, the bandincludes one or more sensors(e.g., NIR optical sensor(s), sensors, etc.), at least one lens, and one or more LEDs, all of which are depicted on an inner side of the band. The devicemay sense user physiological signals on the volar side of the same wrist on which the band is worn.

15 FIG.B 1550 1550 1552 1502 1552 1552 1550 1552 1550 illustrates an example perspective view of another wearable devicefor monitoring glucose and user behavior over time. The wearable devicemay include a bandand an optional smart watch face (not shown). The bandmay be communicatively coupled to the smart watch face (not shown) to communicate data for display on the watch face or on a display (not shown) formed within and/or connected to the band. The bandmay include electronics for monitoring glucose for a user wearing device. In addition, the bandmay include one or more processors coupled to hardware, sensors, or devices for monitoring user behavior, receiving user inputs, and/or generating/displaying recommendations for modifying such user behavior. In some implementations, the deviceis communicatively coupled to one or more other computers, smart phones, servers, ML models, or the like.

1552 1556 1558 1560 1552 1550 15 FIG.C 12 FIGS.C 12 FIG.D In this example, the bandincludes one or more sensors, at least one lens, and one or more LEDs, all of which are depicted on an outer side of the band. The devicemay sense user physiological signals on the volar side of the opposite wrist on which the band is worn, as shown inand/or as described inand/or.

15 FIG.C 15 FIG.B 15 FIG.B 1550 1570 1550 1556 1558 1560 1550 1550 illustrates an example placement of the wearable deviceof. In this example, a useris about to place the volar side of the opposite wrist in which the deviceis being worn against or substantially near to one or more sensors (e.g., one or more sensors, at least one lens, and one or more LEDsshown in) of device. Once placed against or substantially near to such sensors, the wearable devicemay capture one or more measurements (e.g., glucose monitoring measurements, temperature measurements, etc.) as described elsewhere herein.

16 FIG. 1600 1600 1600 illustrates a flow diagram of an example processfor managing blood glucose levels in a subject. The processfunctions to ensure that a user may be educated, assisted, supported, or otherwise motivated to modify a health condition and/or behavior to impact overall health. In some embodiments, the processmay be executed by or using a lifestyle AI coach that provides personalized health recommendations, with a focus on glucose management.

1602 1600 1600 200 At block, the processincludes receiving a plurality of behavioral events. Each behavior event may be correlated to a timestamp within a time period. For example, these events may include blood glucose events, meals consumed, exercise sessions, sleep patterns, and/or stress events, each tagged with a specific time. The timestamp may be used to further assess and/or otherwise determine correlations between event timestamps/events and blood glucose changes. In some embodiments, the processmay include monitoring behavioral events and glucose data over a time period. In such examples, the systemmay capture both explicit user inputs and physiological measurements that may indicate elevated glucose levels even when symptoms are not present or noticeable to the user.

1604 1600 102 200 200 At block, the processincludes monitoring a plurality of images of a body site captured over the time period. For example, the wearable devicemay capture images of a wrist at regular intervals throughout a day. In some embodiments, the image capture may instead occur for another body site in which a view of a venous structure portion may be detected by an image sensor. For example, the systemmay utilize an AI coach or LLM to analyze images for determining glucose patterns and behavioral data to identify rhythms and trends in blood glucose changes. This analysis may enable detection of elevated glucose conditions that typically lack strong symptoms and are not detectable with conventional systems. This may allow the systemto inform users of potential glucose abnormalities before clinical testing reveals prediabetes, hypoglycemia, hyperglycemia, or the like.

1606 1600 4 11 FIGS.A-B At block, the processincludes detecting, within each of the plurality of images, an optical characteristic associated with the body site. This may involve analyzing the darkness, size, or volume of veins visible in the captured images, as described in relation to. In some embodiments, the detecting may further include other assessments such as a morphometric parameter of the venous structure.

1608 1600 200 104 At block, the processincludes determining, for each image and based on the detected optical characteristics, a blood glucose estimation. For example, the systemmay use ML modelsto map the visual changes in one or more venous structures to specific glucose levels.

1610 1600 200 200 At block, the processincludes identifying a pattern in the determined blood glucose estimations. For example, the systemmay detect or recognize trends such as post-meal spikes or overnight fasting levels based on a particular pattern identified in one or more blood glucose estimation. In some embodiments, the systemmay identify patterns in one or more determined blood glucose estimation based on patterns detected or entered by the user including, but not limited to dietary eating patterns, food selection patterns, drug or alcohol consumption patterns, exercise or activity patterns, sleep patterns, stress patterns, and patterns of risk exhibited by the user.

1612 1600 200 At block, the processincludes correlating at least one behavioral event in the plurality of behavioral events to at least one of the identified patterns. For example, the systemmay correlate a sharp increase in glucose levels for the user to a particular meal or a gradual decrease to an exercise session.

1614 1600 200 At block, the processincludes generating, based on the correlating, personalized content for the subject. For example, the systemmay include tailored dietary recommendations, exercise suggestions, or lifestyle modifications as described in paragraphs elsewhere herein to motivate the user to modify a behavior. Example lifestyle modifications may include meal timing, food ingestion order within meals, optimizing sleep duration, routine stress management in response to daily stressors, encouraging social interaction to help enhance motivation dependent on changes in diet, exercise, mood, etc.

1616 1600 200 At block, the processincludes generating one or more nudges for the subject based on the personalized content. The one or more nudges may include prompts or suggestions aimed at encouraging adherence to the personalized content. For example, the systemmay send a reminder to take a walk after a meal that typically causes a glucose spike.

1618 1600 200 102 102 110 108 At block, the processincludes causing presentation of the personalized content and the one or more nudges to the subject. For example, the systemmay generate and present content on the wearable deviceor a device coupled to wearable device. In some embodiments, the information may be sent as a notification to a paired smartphone, one or more buddy devices, and/or an AI coach.

1600 1600 1600 1600 200 200 104 104 200 a In some embodiments, the processmay represent a method of treatment for treating a subject having reduced insulin glycemic control. In some embodiments, the processmay represent a method of treatment for treating a subject having poor insulin glycemic control. In some embodiments, the processmay represent a method of treatment for treating a subject to prevent metabolic disease. In some embodiments, the processmay represent a method of treatment for a subject when blood sugar symptoms are lacking detection. For example, the systemmay determine blood glucose fluctuations when symptoms are not present for the subject. The systemmay capture glucose data in a noninvasive manner and may assess rhythms of blood glucose change (e.g., using ML models,, etc.) and in response, may inform the user even before clinical testing reveals abnormalities. The systemmay make the user aware of the full implications and down the line consequences of developing diabetes.

1600 200 108 104 200 a In some embodiments, the processmay include assessing psychological aspects of the behavior of a user. For example, the systemmay use coachand/or ML modelsto detect or determine default bias patterns that may influence health-related decision making. The systemmay also detect or determine cognitive biases based on user behavior that could lead to poor health choices or resistance to behavioral changes. Such determinations may be used to generate particular nudges and/or content for the user that account for cognitive biases. In some embodiments, the personalized content may include personalized educational content that includes diagrams and text for making users aware of the full implications and long-term consequences of developing prediabetes, diabetes, and/or maintaining elevated glucose levels. This content may be tailored to overcome identified psychological biases and present information in a manner that resonates with the individual user's cognitive patterns.

200 200 In some embodiments, identifying behavioral patterns that correlate with higher risk of glucose spikes may be based on analyzed user data and user history. The systemmay recognize specific activities, meal choices, timing patterns, or lifestyle factors that may precede glucose elevation events. In some embodiments, the systemmay present the personalized content, educational materials, and/or behavioral nudges to the user through a user interface, with timing and presentation methods optimized based on glucose patterns and psychological receptivity of the user.

200 204 200 200 In a non-limiting example, a primary care provider may conventionally have an annual preventive visit with a patient who has prediabetes, elevated blood pressure and rising cholesterol. These are all conditions which benefit from lifestyle change, rather than pharmacologic treatment. Dietary changes, exercise goals, and motivation for behavior change is critical for overcoming these conditions, and the goals and quick strategies are discussed at a primary care visit, but there is no personalized plan of action immediately available to the patient. In this conventional model, there is no monitoring of this behavior and its change until the next annual preventive care visit, unless the patient comes in again for a visit for another issue. This lack of interaction between provider and patient lacks longitudinal support for behavior change. Thus, a year goes by without significant changes to health and without the chance to reverse health risks before the diagnosis of disease occurs. The systemmay provide lifestyle goal setting and monitoring for the user. For example, after a patient exits a primary care visit, an app (e.g., appor other app accessible on any computing device) may be provided to this patient. The patient may input or import health conditions, and in response, the app will pre-populate goals for diet (e.g., caloric intake and beneficial/adverse food groups to monitor for intake), exercise (e.g., minutes per week of aerobic and strength training), which will form the content of history, context, and prompts provided to the LLM. Using this history and low-burden user-provided demographic, laboratory, habits (e.g., diet preferences, work schedule, preferred exercises, etc.) data, the app may use systemto provide scheduled nudges to begin 1-2 weekly behavior changes. These behavior changes will progressively update based on achieving goals. The systemmay utilize natural milestones (e.g., the beginning of a school/work year, birthdays, vacations, etc.) to align goals for the year. In some embodiments, guidance may be automated for a set time period (e.g., 3 months, 6 months, 1 year, etc.) and may result in a suggestion for repeat testing and a primary care provider visit at the end of this time period. The app can track progress on these lifestyle changes using a combination of user-reported and passive data collected by mobile device health apps, wearable device data, etc.

1600 200 102 102 108 quinoa In some embodiments, the processmay further include correlating the contents of a last consumed meal to the determined blood glucose estimation, and generating feedback on the last consumed meal as part of the personalized content. For example, the systemmay receive an indication from a user of wearable devicethat the last consumed meal included white rice, teriyaki chicken, and a soda. The wearable devicemay detect a glucose response at some time after the meal. For example, a glucose response may be detected as 185 mg/dL indicating a postprandial spike. The lifestyle AI coachmay assess the detected glucose response and the meal and may provide feedback. Example feedback for this example may include a message or indication such as ‘your blood sugar spiked higher than expected after your last meal. The white rice and soda are likely contributors due to having a high glycemic index. For a more stable blood sugar response, try substituting white rice with brown rice or, and replace soda with water or unsweetened tea.’

200 102 102 108 quinoa In another example, the systemmay receive an indication from a user of wearable devicethat the last consumed meal included grilled salmon,, and steamed broccoli. The wearable devicemay detect a glucose response at some time after the meal. For example, a glucose response may be detected as 110 mg/dL, which may be determined to be within a target postprandial range. The lifestyle AI coachmay assess the detected glucose response and the meal and may provide feedback. Example feedback for this example may include a message or indication such as ‘Great job! Your blood glucose stayed within a healthy range after your meal. The protein and fiber from salmon and vegetables likely helped slow digestion and glucose absorption. Keep up the balanced meals.’

200 102 108 In another example, the systemmay receive an indication from the user that the last consumed items were a granola bar and a fruit smoothie. The wearable devicemay detect a glucose response of approximately 165 mg/dL after consumption. The lifestyle AI coachmay assess this elevated response and correlate it with the reported intake. Example feedback may include an indication such as ‘although your snack choices seemed healthy, the granola bar and smoothie may have had added sugars. Try checking the labels for total sugar content and opt for whole fruit and nuts instead next time.’

200 102 200 108 109 In yet another example, the systemmay detect that a user consumed cheese pizza at approximately 10:30 PM. A fasting glucose level measured by the wearable devicethe following morning may indicate an elevated blood glucose response value of approximately 125 mg/dL. The system(e.g., lifestyle AI coachor agent) may evaluate the temporal relationship between the meal and the glucose reading and provide feedback such as ‘your fasting blood sugar was elevated this morning, which may be linked to your late-night meal. Try to finish eating 2-3 hours before bedtime to help your body stabilize glucose levels overnight.’

200 229 102 108 In a further example, the systemmay receive input that the user consumed a chicken wrap and an apple. Additionally, activity sensor data (e.g., from one or more sensors) may indicate that the user engaged in 30 minutes of walking shortly after the meal. The wearable devicemay detect a mild glucose elevation to approximately 98 mg/dL with a quick return to baseline levels. The lifestyle AI coachmay correlate the physical activity with the favorable glucose trend and provide feedback such as ‘Excellent response! Light activity after meals can improve your body's glucose handling. Your walk after lunch helped minimize your blood sugar increase. Consider making this a regular habit.’

1600 200 The processmay also include generating feedback that may suggest timing for a next meal and/or educational content. The educational content may include a meal plan for a user to follow as a way to minimize glucose spikes over a second time period. The second time period may be determined and/or otherwise selected by systembased at least in part on the blood glucose estimations determined for the user.

1600 1600 1600 1600 200 200 In some embodiments, the processmay further include monitoring a behavior or input responsive to the presentation of the one or more nudges. For example, food input, exercise input, and/or detected blood glucose levels may be monitored over time. The processmay iteratively repeat the monitoring of additional images of the body site, for example to determine updated blood glucose levels for the user. The processmay then update the personalized content and nudges based on any detected additional inputs, behavioral events, and images. In some embodiments, the processmay also include adapting the personalized content based on contextual information associated with the subject, such as current location, time of day, upcoming events, or recent activities. For example, if the systemnudged the user to take a walk after a meal and the user complied (according to the monitoring) and the walk resulted in improved glucose levels, the systemmay reinforce this behavior in future recommendations.

1600 108 102 102 102 108 200 108 In some embodiments, the processmay include determining a nudge mode from a plurality of nudge modes based on the plurality of behavioral events and the monitoring of the plurality of images, and generating the one or more nudges according to the determined nudge mode. In some embodiments, the nudge mode may represent an indication of how many nudges may be provided to a user over a time period. In some embodiments, the nudge mode may represent which words to use in the nudges where a light mode may provide suggestions or general guidelines, a heavy mode may provide rules to abide by and/or explicit meal plans, exercises, and/or timing of both. The nudge mode may be determined by the AI coach, the wearable device, or may be user selected at device(or a companion device coupled to device). In some embodiments, nudge modes may also be defined by a tone and/or approach of the lifestyle AI coach. The tone and/or approach may be defined for each nudge mode and may be dynamically adjusted based on user feedback. For example, the systemmay implement a quick poll using emojis to gauge a current emotional state of the user, which may trigger the AI coachto generate empathetic or assertive approaches, accordingly.

110 In a non-limiting example, if a user links a support family member or friend (e.g., buddy device) and notes messages between the two individuals, the electronic nudge may be skipped in favor of the personal interaction between the two individuals. The nudge mode may then adapt to suggest this personal interaction at regular intervals in the future.

1600 1600 200 200 In some embodiments, the processmay further include determining a glucose time in range for a user based on the blood glucose estimations determined for the user over time. The processmay then generate a time to trigger the presentation of the one or more nudges based on the glucose time in range. For example, if the determined glucose blood levels are consistently within range, the systemmay reduce the frequency of nudges to weekly check-ins. Similarly, if the detected blood glucose levels frequently fall out of range in the evenings, the systemmay increase the frequency of evening nudges to encourage healthier choices.

1600 200 In some embodiments, the processmay include generating a grocery shopping list curated according to the blood glucose estimations, and may automatically place items from the list in an online shopping cart and/or may place an order for the items. For example, if the glucose levels for a user indicate a lack of or a reduced amount of fiber as compared to a predefined fiber goal, the systemmay generate a shopping list of high-fiber foods and may also place an order for delivery or pickup of the proposed items.

102 108 110 The systems and methods described herein and variations thereof can be embodied and/or implemented at least in part as a machine configured to receive a computer-readable medium storing computer-readable instructions. The instructions may be executed by computer-executable components integrated with the system and one or more portions of the processor on the assemblies described herein and/or computing devices,,, etc. The computer-readable medium can be stored on any suitable computer-readable media such as RAMs, ROMs, flash memory, EEPROMs, optical devices (e.g., CD or DVD), hard drives, floppy drives, or any suitable device. The computer-executable component may include any suitable dedicated hardware or hardware/firmware combination that can alternatively or additionally execute the instructions.

17 17 FIGS.A-B 17 FIG.A 1700 1700 200 109 108 200 illustrate an example user interfaceof a conversational exchange depicting an interaction sequence between a user and an AI-driven user interface.illustrates a user interfacethat depicts content and prompts that function to guide a user to a personalized set of healthy grocery recommendations at stores proximate to the user. The recommendations may be provided through an adaptive prompt-response framework, such as system(e.g., agentor lifestyle AI coach). Each AI output is informed by structured logic layers, including intent classification, context acquisition, location-based filtering, and affordability heuristics. The systemmay improve user experience in real-time grocery planning scenario and may be extended to restaurant recommendations, nutrition analysis, and/or digital couponing applications.

102 200 200 1702 200 200 1704 A user of wearable device, for example, may initiate a session by transmitting a general query, such as “hello.” In response, the systemanalyzes the user input and generates a context-relevant follow-up query, aimed at refining the scope of assistance. The system,may prompt the user with a disambiguating question, such as ‘Are you looking for healthy items in a nearby grocery store or healthy options on the menu of a specific restaurant?’ This step initiates the intent detection module (not shown) of the system, which classifies the query under a grocery-focused information retrieval context when the user responds with the keyword grocery. The systemmay then proceed to refine the inquiry by soliciting additional contextual information, including geographic and economic constraints. This stage engages the context enrichment engine (not shown), which may collect geospatial and budgetary parameters to support localized and personalized recommendations.

200 94027 200 200 200 1 3 200 200 Upon the user providing a zip code and a budget, the systemmay generate personalized grocery store recommendations. In this example, the user provides two inputs: a zip code () and a budget ($97). These inputs are parsed and tokenized by the system. Subsequently, the systemaccesses a database of grocery stores mapped to the provided zip code. Based on criteria including distance, customer ratings, price level, and product availability, the systemreturns a structured list of top grocery options. This output is presented in a ranked and enumerated format (e.g.,to) to facilitate user comprehension and decision-making. The formatting and prioritization logic is driven by a multi-criteria scoring algorithm, which weighs the importance of review ratings, affordability, and physical proximity to the zip code centroid. In some embodiments, the systemmay further refine or filter store recommendations based on the provided budget (e.g., $97). While the illustrated response does not explicitly tie pricing to individual product suggestions, in additional embodiments, the systemmay interface with retailer APIs to assess whether recommended stores offer a sufficient quantity of healthy foods (e.g., fresh produce) within a particular declared budget.

17 FIG.B 1710 1700 1710 200 200 1712 illustrates a user interfacedepicting a continuation of the user interface. User interfaceincludes further interaction between the user and the AI-driven system. For example, after the user selects a grocery store from the provided options, the systemproceeds to offer personalized healthy food recommendations.

200 200 200 1710 200 17 FIG.C 17 FIG.D 17 FIG.B The user responds to the grocery store recommendations by selecting ‘Trader Joe's’. The systemacknowledges this choice and initiates the next phase of assistance, focusing on healthy food recommendations. The systemmay prompt the user with additional questions, such as: ‘Great choice! What kind of healthy foods are you interested in? (e.g., fruits, vegetables, lean proteins, whole grains)’. This question serves to narrow down user preferences and dietary interests. Other questions are possible. The systemprocesses any user input responses and may generate a tailored list of healthy fruit and vegetable recommendations, as shown inand. The recommendations are presented in a structured, easy-to-read format, with each item accompanied by its potential health benefits. The user interfaceinshowcases the ability of systemto provide personalized, context-aware recommendations while maintaining an engaging conversation flow. The AI-driven interface adapts to user responses, offering relevant information and anticipating potential next steps in a healthy grocery shopping journey for the user.

17 FIG.C 1720 1700 1710 200 200 1722 illustrates a user interface, which represents a further continuation of the user interfaceand, showing the ongoing interaction between the user and the AI-driven system. After the user expresses interest in a week of healthy food items at Trader Joe's, the systemprovides the list. Each element in the list includes the name of the food, a price, and an educational portion (e.g., portion) to teach the user about the food item and how that item may impact the human body.

17 FIG.D 1730 1700 1710 1720 1732 1734 200 1736 1730 200 1700 1710 1720 1730 200 200 200 200 depicts a user interfaceillustrating a final part of this conversational exchange of the user interfaces,, and, showcasing a capability to provide additional health-related information and wrap up the interaction with additional food items and educational content (e.g., content) and a total cost. The systemmay provide a summary of how the food item combination provides nutrition and fits within a predefined budget (e.g., content). In some embodiments, the user interfacemay further include prompts to check if the user wishes to add items to an electronic/online shopping cart and/or place an order at Trader Joe's online for the proposed items. If the user requests to fill an online shopping cart, the systemmay do so and add the items to an electronic shopping cart for the selected store and items indicated in the output of user interfaces,,, and. If the user requests to make the purchase, the systemmay function to fill the online shopping cart and present payment options to the user. In some embodiments, the systemmay perform the purchase, for example, if the user preconfigured payment options and automated cart purchases. In some embodiments, the systemmay add items suggested for purchase to an existing electronic shopping cart. This functionality demonstrates the ability of systemto integrate recommendations with practical shopping assistance.

18 18 FIGS.A-B 1800 200 109 108 illustrate example user interfacesof a conversational exchange depicting an interaction sequence between a user and an AI-driven user interface for providing personalized restaurant based recommendations. The recommendations may be provided through an adaptive prompt-response framework, such as system(e.g., agentor lifestyle AI coach).

18 FIG.A 1800 200 1802 1804 1806 1808 illustrates an example user interfaceof a conversation about a restaurant the user wishes to visit. In this example, the systemprovided at the request of the user, a list of health options at a McDonald's restaurant. Each item lists calories, sugar content, carbohydrate content, and an educational description.

18 FIG.B 1810 200 1812 1814 200 illustrates an example user interfaceof a conversation about another restaurant the user wishes to visit. In this example, the systemprovided at the request of the user, a list of health options at a Starbucks restaurant. Each item lists calories, sugar content, carbohydrate content, and an educational description. Additional educational help and/or suggestionsmay be provided based on user goals and/or blood glucose levels known to system.

19 FIG. 1900 1900 1900 1900 218 1900 108 104 104 a b illustrates a flow diagram of an example processfor managing behavior in a subject (e.g., a user). The processfunctions to ensure that a user may be educated, assisted, supported, or otherwise motivated to modify a health condition and/or behavior to impact overall health. In some embodiments, the processmay be executed by or using a lifestyle AI coach that provides personalized health recommendations, with a focus on particular behaviors or patterns of behaviors. In some embodiments, processmay be executed by a processor, such as processor. In some embodiments, processmay be executed by a processor and receive additional data from AI coach (e.g., coach) and/or other ML model (e.g., ML model, ML model, or the like).

1902 1900 200 200 228 229 102 102 102 108 200 108 At block, the processincludes monitoring a plurality of behavioral events. Each event may be correlated to a timestamp within a time period. For example, each event may occur at the timestamped time. In some embodiments, each event may be entered or detected by one or more components of the system. In some embodiments, the systemmay monitor particular behaviors using one or more sensorsand/oron wearable device. In some embodiments, the systemmay monitor the behaviors using other devices communicatively coupled to deviceand/or coach. In some embodiments, the systemmay receive behavioral events as data from a third party, a user, a lifestyle coachor the like.

200 200 3 For example, the systemmay monitor for or receive behavior and/or behavior events corresponding to a user. The behaviors and/or behavior events may include meals consumed, exercise sessions, sleep patterns, stress events, or other behaviors described elsewhere herein. Each behavior or behavior event may be tagged with specific timestamps. For example, the systemmay track naan and paneer consumed at a specific time, orwheat rolls and 3/4 cup okra eaten at another timestamp.

200 200 108 In some embodiments, the monitoring may include monitoring or receiving blood glucose data for the subject. Identifying the pattern in the plurality of behavioral events may include identifying a glucose time in range for the subject. The system may use such data to select, based on the glucose time in range, one or more times in which to present the one or more nudges to the subject. For example, the systemmay adjust frequency based on glucose control. If the time in range is consistently high, the systemmay reduce monitoring frequency to weekly checks. However, if values begin to fall out of a predefined range more often, the AI coach, for example, may adapt by deploying more frequent reminders or nudges.

1904 200 200 200 108 200 104 At block, the system identifies a pattern in the plurality of behavioral events. For example, the systemmay analyze the collected behavioral data and detect recurring trends, correlations, or sequences that may impact the health of the user, such as glucose regulation, food intake, etc., The pattern identification may reveal subtle behavioral rhythms that precede glucose spikes or other health-related events, even when clinical symptoms are not yet apparent to the subject. For example, the systemmay detect trends such as post-meal spikes or overnight fasting levels based on a particular pattern identified in one or more blood glucose estimation. In some embodiments, the systemmay identify when consistent nighttime glucose spikes are detected for a user during Ramadan, for example. In response, the AI coachmay adjust recommendations and shift to a light mode of nudging until the end of the fasting period. In some embodiments, the systememploys ML modelsto analyze changes over time and assess blood glucose or other past behaviors to infer future behaviors for the user.

200 200 200 200 In some embodiments, the systemmay correlate multiple types of behavioral events with physiological measurements. For example, the systemmay determine temporal correlations by tracking frequency and timing of meals and analyzing glucose spike patterns to identify relationships between eating behaviors and physiological responses. The systemmay also integrate contextual data. For example, the systemmay process mobile device-type measurements or wearable device measurements including frequency of phone/app interaction and patterns of use, frequency and pattern of communication via text message, email message, phone call, or voice message, and/or location services accessed during particular time periods (e.g., restaurant, game, work, home, vacation).

200 In some embodiments, the systemmay identify complex behavioral patterns by correlating multiple data streams, applying machine learning algorithms, and continuously refining pattern recognition through iterative analysis of user responses and physiological feedback.

1906 1900 200 200 At block, the processincludes generating personalized content for the subject based on the identified pattern. The personalized content may include one or more of: dietary recommendations, exercise suggestions, and lifestyle modifications. In some embodiments, at least one of the plurality of behavioral events may include a last consumed meal and the personalized content may include feedback on the last consumed meal. The feedback may include suggested timing for a next meal and educational content. The educational content may include a meal plan for the subject that minimizes glucose spikes over a second time period. For example, if the systemdetermines behavior trending toward increased blood glucose levels, the systemmay refine guidance or feedback to prompt the user to consume protein-containing foods first at a next meal.

200 200 200 200 200 200 200 quinoa The systemmay generate and tailor recommendations in the form of personalized content for the subject. The personalized content may provide tailored feedback and recommendations based on meal intake, detected glucose response, and/or contextual behaviors of the user. For example, the systemmay assess glucose levels for the user following a specific meal and may generate interpretive feedback indicating how the contents of the meal (e.g., such as high glycemic index foods) contributed to a postprandial spike. This feedback may include suggestions for alternative food options, such as substituting white rice with brown rice or, or replacing soda with water or unsweetened beverages, thereby guiding the user toward more stable glucose responses. When the systemdetects a favorable glucose response, positive reinforcement messages may be generated that commend the user for a well-balanced meal, encouraging continued adherence to healthy dietary habits. In cases where meals perceived as healthy produce unexpected glucose elevations, the systemmay provide explanatory content highlighting the potential presence of added sugars or refined ingredients, coupled with tips for improved label reading or more nutrient-dense alternatives. The systemmay also account for behavioral context such as the timing of a meal, and may further offer temporal advice (e.g., avoid late-night eating to improve fasting glucose outcomes). Moreover, when physical activity is detected post-meal and correlates with a favorable glucose trend, the systemmay reinforce this behavior by informing the user that the performed walk helped minimize a blood sugar increase and nudges may be scheduled to suggest the walk be repeated regularly. Through these dynamic, context-sensitive interactions, the systemmay deliver personalized coaching that adapts to the unique physiology, habits, and preferences of the user to promote improved glycemic control and long-term health outcomes.

1908 1900 200 200 At block, the processincludes one or more nudges for the subject based on the personalized content. The one or more nudges may include prompts or suggestions aimed at encouraging adherence to executing suggested behavior changes provided in the personalized content. For example, one or more nudges may be generated and provided by systemto indicate that the user should attempt to decrease refined carbohydrates and increase high fiber and protein in each meal responsive to detecting a pattern indicating that post-prandial glucose is continually highly elevated after meals. The systemmay continue with another nudge if the first nudge did not motivate the user to modify the meals accordingly. The second nudge may include suggestions about performing post-prandial light exercise with timers and nudges towards 30 minutes of moderate intensity exercise per day.

200 200 In another example, the systemmay generate a prompt that includes a nudge warning against drinking more than two drinks, for example, if the user is at a bar with friends at night, as this can cause a significant glucose spike. In yet another example, the systemmay generate a prompt that includes a nudge that preemptively suggests foods high in fiber and protein in response to determining prior habits or patterns for the user resulted in over-eating and high glucose when a user is home alone at a particular time.

1910 200 200 204 200 110 108 200 102 102 110 108 At block: The systemcauses presentation of the personalized content and the one or more nudges to the subject. For example, the systemmay generate content and/or user interfaces and present such content in the user interfaces, browsers, and/or apps. In some embodiments, the systemmay generate and present the personalized content through buddy devices, lifestyle AI coach, or another interface associated with the user. In general, the systemmay present the personalized content and/or nudges through various formats including, but not limited to: short games, popups, text messages, emails, summary screens, alerts based on blood glucose levels, prompts for logging diet intake, etc. The presentation may occur on the wearable deviceor another device communicatively coupled to wearable device. In some embodiments, the presentation may occur or be sent as a notification to a paired smartphone, one or more buddy devices, and/or an AI coach.

1900 1900 1900 1900 200 200 104 104 200 a In some embodiments, the processmay represent a method of treatment for treating a subject having reduced insulin glycemic control as assessed by behavioral analysis. In some embodiments, the processmay represent a method of treatment for treating a subject having poor insulin glycemic control. In some embodiments, the processmay represent a method of treatment for treating a subject to prevent metabolic disease by modifying behaviors over time. In some embodiments, the processmay represent a method of treatment for a subject when blood sugar symptoms are lacking detection. For example, the systemmay determine blood glucose fluctuations when symptoms are not present for the subject. The systemmay capture behavioral data in a noninvasive manner and may assess rhythms of behavior (e.g., using ML models,, etc.) and in response, may inform the user even before clinical testing reveals abnormalities. The systemmay make the user aware of the full implications and down the line consequences of developing prediabetes and/or diabetes.

1900 210 209 209 210 1900 102 204 a b In some embodiments, the operations of processmay be performed by an artificial intelligence computer agent (e.g., predictive systemand/or generative system, generative systemand/or prediction system, etc.) that may use data generated by monitoring the behavior of the subject and a number of template guides to generate the one or more nudges. Such nudges may be used to coach the subject to modify the behavior over an upcoming time period. In some embodiments, the operations of processare carried out locally on wearable deviceby an app.

1900 200 200 In some embodiments, the processmay further include monitoring a behavior or input responsive to the presentation of the one or more nudges and iteratively repeating the monitoring of behavior exhibited by the subject. In response to detecting additional behavioral events, the systemmay identify an additional pattern in the additional behavioral events and may update the personalized content according to the additional behavioral events based on the identified additional pattern. The systemmay generate an updated nudge corresponding to the updated personalized content and may cause presentation of the updated nudge and the updated personalized content to the subject.

200 200 200 200 108 200 For example, the systemmay track user compliance, such as monitoring the user over time. If the systemnudged the user to take a walk after a meal and the user complied (according to the monitoring) and the walk resulted in improved glucose levels, the systemmay reinforce this behavior in future recommendations. Similarly, the systemmay adapt based on new data. For example, if the lifestyle AI coachdetects systematic elevations in glucose levels over the course of a month for the user, the systemmay provide nudges reminding the user of healthful meal components and encourage the reintroduction of routine exercise.

1900 1 s The processmay include determining a historical adherence to previous nudges or recommendations generated and/or presented to the user. The historical adherence may be used to select a nudge mode from a plurality of nudge modes. Example nudge modes may include a light mode, a regular mode, and a heavy mode. The light mode may be activated during at least one of: holidays, travel, postpartum periods, short-term illnesses, or user-designated cheat days. The heavy mode may be activated during at least one of: periods following discontinuation of one or more medications (e.g., GLP-, supplements, insulin, etc.) or when the subject sets aggressive health goals. The regular mode may represent a mode between the light mode and the heavy mode. In general, the one or more nudges may be generated according to a determined (or otherwise configured) nudge mode.

1900 200 108 The processmay further include adapting the personalized content based on contextual information associated with the subject. The contextual information may include at least one of: a current location, a time of day, one or more upcoming events, or one or more recent activities. In general, the systemmay adapt based on circumstances of the user. For example, if a user is attending a conference and unable to adhere to typical recommendations, the lifestyle AI coachmay inquire about the conference details and adjust a nudging mode or strategy, accordingly, thereby resuming normal (e.g., regular, standard mode) recommendations after the conference concludes.

1900 In some embodiments, the personalized content described herein may include a grocery shopping list curated according to the identified pattern. The processmay use such information to further perform automatic placement of a plurality of items on the grocery shopping list in an online shopping cart or automatic placement of an order of the plurality of items on the grocery shopping list.

200 1900 17 18 FIGS.A-B In some embodiments, the personalized content described herein may be presented as one or more venues and a grocery shopping list curated according to the identified pattern, a budget, a location of the subject, and a determination of an availability of a plurality of items on the grocery shopping list. For example, the systemmay determine, based on user location and budget, which venues (e.g., grocery stores) to present to the user according to the location, the budget, and a determination of availability of health-based food and particular listed grocery items on the shopping list. The processmay further include automatically placing the plurality of items on the grocery shopping list in an online shopping cart or automatically placing an order of the plurality of items on the grocery shopping list, as shown in further detail in.

200 200 200 200 In a non-limiting example, one or more grocery store venues may be selected and suggested by systembased on one or more of: a proximity to the user, user budget, and availability of healthy items. In some embodiments, a grocery list may be automated based on the systemdetermining a venue and a day of the week. The systemmay perform a web-scrape to provide current products, ingredients, and pricing available at a venue of interest. The systemcan read and interpret product labels. Thus, the shopping list can be curated according to the identified user patterns and also a determined availability of the products on the list specific to the selected one or more venues.

In some embodiments, the personalized content may include a plurality of restaurant menu suggestions and food data corresponding to each of the plurality of restaurant menu suggestions. The food data may indicate one or more of: calorie content, sugar content, carbohydrate content, and at least one reason for selecting the respective restaurant menu suggestions. In some embodiments, selection of the restaurant may be based on one or more of: identified behavior patterns for the user, user location or proximity, user budget, and availability of healthy items on the menu.

200 1600 200 In a non-limiting example, one or more restaurant venues may be selected and suggested by systembased on one or more of: a proximity to the user, user budget, and availability of healthy items. Menu items may also be suggested. The processmay further include automatically placing the suggested items form the menu in an online shopping cart for purchasing restaurant food. The systemmay also automatically perform a purchase of the items in the online shopping cart according to user rules and/or requests.

1900 108 110 In some embodiments, the processmay be performed by an artificial intelligence health coach (e.g., AI coach) that may be configured to share data with a buddy computing device (e.g., device) according to permissions associated with the subject. The shared data may include one or more of: a blood glucose estimation, identified behavioral patterns, a status update corresponding to the subject, and a summary corresponding to the subject. Such data may be multi-modal including, but not limited to one or more of: blood glucose graphs, weights, language or nudges, or the like.

110 102 200 The buddy computing devicedescribed herein may be used with wearable device, a CGM or BGM, or as a standalone computing device enrolled as a trusted member with one or more components of system.

The systems and methods described herein may be implemented as a mobile application, wearable-integrated platform, or web-based interface. The LLMs/ML models/AI may be integrated with external application programming interfaces or agents to access nutritional databases, monitor physical activity data, or synthesize custom behavioral strategies tailored to individual users' progress and lifestyle constraints.

The term “about” or “approximately,” when used before a numerical designation or range (e.g., to define a length or pressure), indicates approximations which may vary by (+) or (−) 5%, 1% or 0.1%. All numerical ranges provided herein are inclusive of the stated start and end numbers. The term “substantially” indicates mostly (i.e., greater than 50%) or essentially all of a device, substance, or composition.

As used herein, the term “comprising” or “comprises” is intended to mean that the devices, systems, and methods include the recited elements, and may additionally include any other elements. “Consisting essentially of” shall mean that the devices, systems, and methods include the recited elements and exclude other elements of essential significance to the combination for the stated purpose. Thus, a system or method consisting essentially of the elements as defined herein would not exclude other materials, features, or steps that do not materially affect the basic and novel characteristic(s) of the claimed disclosure. “Consisting of” shall mean that the devices, systems, and methods include the recited elements and exclude anything more than a trivial or inconsequential element or step. Embodiments defined by each of these transitional terms are within the scope of this disclosure.

The systems and methods of the embodiments described herein and variations thereof can be embodied and/or implemented at least in part as a machine for receiving a computer-readable medium storing computer-readable instructions. The instructions may be executed by computer-executable components integrated with the system and one or more portions of the processor on the wearable device and/or computing device. In some embodiments, the instructions may be executed by computer-executable components that are in communication with the wearable device. The computer-readable medium can be stored on any suitable computer-readable media such as RAMs, ROMs, flash memory, EEPROMs, optical devices (e.g., an image or DVD), hard drives, floppy drives, or any suitable device. The computer-executable component may be a general or application-specific processor, but any suitable dedicated hardware or hardware/firmware combination can alternatively or additionally execute the instructions.

As used in the description and claims, the singular form “a”, “an” and “the” include both singular and plural references unless the context clearly dictates otherwise. For example, the term “sensor” may include, and is contemplated to include, a plurality of sensors. At times, the claims and disclosure may include terms such as “a plurality,” “one or more,” or “at least one;” however, the absence of such terms is not intended to mean, and should not be interpreted to mean, that a plurality is not conceived.

The examples and illustrations included herein show, by way of illustration and not of limitation, specific embodiments in which the subject matter may be practiced. Other embodiments may be utilized and derived therefrom, such that structural and logical substitutions and changes may be made without departing from the scope of this disclosure. Such embodiments of the inventive subject matter may be referred to herein individually or collectively by the term “invention” merely for convenience and without intending to voluntarily limit the scope of this application to any single invention or inventive concept, if more than one is in fact disclosed. Thus, although specific embodiments have been illustrated and described herein, any arrangement calculated to achieve the same purpose may be substituted for the specific embodiments shown. This disclosure is intended to cover any and all adaptations or variations of various embodiments. Combinations of the above embodiments, and other embodiments not specifically described herein, will be apparent to those of skill in the art upon reviewing the above description.